drivers/net/tun.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  TUN - Universal TUN/TAP device driver.
   4  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   5  *
   6  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
   7  */
   8
   9 /*
  10  *  Changes:
  11  *
  12  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  13  *    Add TUNSETLINK ioctl to set the link encapsulation
  14  *
  15  *  Mark Smith <markzzzsmith@yahoo.com.au>
  16  *    Use eth_random_addr() for tap MAC address.
  17  *
  18  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  19  *    Fixes in packet dropping, queue length setting and queue wakeup.
  20  *    Increased default tx queue length.
  21  *    Added ethtool API.
  22  *    Minor cleanups
  23  *
  24  *  Daniel Podlejski <underley@underley.eu.org>
  25  *    Modifications for 2.3.99-pre5 kernel.
  26  */
  27
  28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  29
  30 #define DRV_NAME        "tun"
  31 #define DRV_VERSION     "1.6"
  32 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  33 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  34
  35 #include <linux/module.h>
  36 #include <linux/errno.h>
  37 #include <linux/kernel.h>
  38 #include <linux/sched/signal.h>
  39 #include <linux/major.h>
  40 #include <linux/slab.h>
  41 #include <linux/poll.h>
  42 #include <linux/fcntl.h>
  43 #include <linux/init.h>
  44 #include <linux/skbuff.h>
  45 #include <linux/netdevice.h>
  46 #include <linux/etherdevice.h>
  47 #include <linux/miscdevice.h>
  48 #include <linux/ethtool.h>
  49 #include <linux/rtnetlink.h>
  50 #include <linux/compat.h>
  51 #include <linux/if.h>
  52 #include <linux/if_arp.h>
  53 #include <linux/if_ether.h>
  54 #include <linux/if_tun.h>
  55 #include <linux/if_vlan.h>
  56 #include <linux/crc32.h>
  57 #include <linux/nsproxy.h>
  58 #include <linux/virtio_net.h>
  59 #include <linux/rcupdate.h>
  60 #include <net/net_namespace.h>
  61 #include <net/netns/generic.h>
  62 #include <net/rtnetlink.h>
  63 #include <net/sock.h>
  64 #include <net/xdp.h>
  65 #include <net/ip_tunnels.h>
  66 #include <linux/seq_file.h>
  67 #include <linux/uio.h>
  68 #include <linux/skb_array.h>
  69 #include <linux/bpf.h>
  70 #include <linux/bpf_trace.h>
  71 #include <linux/mutex.h>
  72 #include <linux/ieee802154.h>
  73 #include <linux/if_ltalk.h>
  74 #include <uapi/linux/if_fddi.h>
  75 #include <uapi/linux/if_hippi.h>
  76 #include <uapi/linux/if_fc.h>
  77 #include <net/ax25.h>
  78 #include <net/rose.h>
  79 #include <net/6lowpan.h>
  80
  81 #include <linux/uaccess.h>
  82 #include <linux/proc_fs.h>
  83
  84 static void tun_default_link_ksettings(struct net_device *dev,
  85                                        struct ethtool_link_ksettings *cmd);
  86
  87 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  88
  89 /* TUN device flags */
  90
  91 /* IFF_ATTACH_QUEUE is never stored in device flags,
  92  * overload it to mean fasync when stored there.
  93  */
  94 #define TUN_FASYNC      IFF_ATTACH_QUEUE
  95 /* High bits in flags field are unused. */
  96 #define TUN_VNET_LE     0x80000000
  97 #define TUN_VNET_BE     0x40000000
  98
  99 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 100                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 101
 102 #define GOODCOPY_LEN 128
 103
 104 #define FLT_EXACT_COUNT 8
 105 struct tap_filter {
 106         unsigned int    count;    /* Number of addrs. Zero means disabled */
 107         u32             mask[2];  /* Mask of the hashed addrs */
 108         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 109 };
 110
 111 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 112  * to max number of VCPUs in guest. */
 113 #define MAX_TAP_QUEUES 256
 114 #define MAX_TAP_FLOWS  4096
 115
 116 #define TUN_FLOW_EXPIRE (3 * HZ)
 117
 118 /* A tun_file connects an open character device to a tuntap netdevice. It
 119  * also contains all socket related structures (except sock_fprog and tap_filter)
 120  * to serve as one transmit queue for tuntap device. The sock_fprog and
 121  * tap_filter were kept in tun_struct since they were used for filtering for the
 122  * netdevice not for a specific queue (at least I didn't see the requirement for
 123  * this).
 124  *
 125  * RCU usage:
 126  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 127  * other can only be read while rcu_read_lock or rtnl_lock is held.
 128  */
 129 struct tun_file {
 130         struct sock sk;
 131         struct socket socket;
 132         struct tun_struct __rcu *tun;
 133         struct fasync_struct *fasync;
 134         /* only used for fasnyc */
 135         unsigned int flags;
 136         union {
 137                 u16 queue_index;
 138                 unsigned int ifindex;
 139         };
 140         struct napi_struct napi;
 141         bool napi_enabled;
 142         bool napi_frags_enabled;
 143         struct mutex napi_mutex;        /* Protects access to the above napi */
 144         struct list_head next;
 145         struct tun_struct *detached;
 146         struct ptr_ring tx_ring;
 147         struct xdp_rxq_info xdp_rxq;
 148 };
 149
 150 struct tun_page {
 151         struct page *page;
 152         int count;
 153 };
 154
 155 struct tun_flow_entry {
 156         struct hlist_node hash_link;
 157         struct rcu_head rcu;
 158         struct tun_struct *tun;
 159
 160         u32 rxhash;
 161         u32 rps_rxhash;
 162         int queue_index;
 163         unsigned long updated ____cacheline_aligned_in_smp;
 164 };
 165
 166 #define TUN_NUM_FLOW_ENTRIES 1024
 167 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 168
 169 struct tun_prog {
 170         struct rcu_head rcu;
 171         struct bpf_prog *prog;
 172 };
 173
 174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 175  * device, socket filter, sndbuf and vnet header size were restore when the
 176  * file were attached to a persist device.
 177  */
 178 struct tun_struct {
 179         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 180         unsigned int            numqueues;
 181         unsigned int            flags;
 182         kuid_t                  owner;
 183         kgid_t                  group;
 184
 185         struct net_device       *dev;
 186         netdev_features_t       set_features;
 187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 188                           NETIF_F_TSO6 | NETIF_F_GSO_UDP_L4)
 189
 190         int                     align;
 191         int                     vnet_hdr_sz;
 192         int                     sndbuf;
 193         struct tap_filter       txflt;
 194         struct sock_fprog       fprog;
 195         /* protected by rtnl lock */
 196         bool                    filter_attached;
 197         u32                     msg_enable;
 198         spinlock_t lock;
 199         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 200         struct timer_list flow_gc_timer;
 201         unsigned long ageing_time;
 202         unsigned int numdisabled;
 203         struct list_head disabled;
 204         void *security;
 205         u32 flow_count;
 206         u32 rx_batched;
 207         atomic_long_t rx_frame_errors;
 208         struct bpf_prog __rcu *xdp_prog;
 209         struct tun_prog __rcu *steering_prog;
 210         struct tun_prog __rcu *filter_prog;
 211         struct ethtool_link_ksettings link_ksettings;
 212         /* init args */
 213         struct file *file;
 214         struct ifreq *ifr;
 215 };
 216
 217 struct veth {
 218         __be16 h_vlan_proto;
 219         __be16 h_vlan_TCI;
 220 };
 221
 222 static void tun_flow_init(struct tun_struct *tun);
 223 static void tun_flow_uninit(struct tun_struct *tun);
 224
 225 static int tun_napi_receive(struct napi_struct *napi, int budget)
 226 {
 227         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 228         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 229         struct sk_buff_head process_queue;
 230         struct sk_buff *skb;
 231         int received = 0;
 232
 233         __skb_queue_head_init(&process_queue);
 234
 235         spin_lock(&queue->lock);
 236         skb_queue_splice_tail_init(queue, &process_queue);
 237         spin_unlock(&queue->lock);
 238
 239         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 240                 napi_gro_receive(napi, skb);
 241                 ++received;
 242         }
 243
 244         if (!skb_queue_empty(&process_queue)) {
 245                 spin_lock(&queue->lock);
 246                 skb_queue_splice(&process_queue, queue);
 247                 spin_unlock(&queue->lock);
 248         }
 249
 250         return received;
 251 }
 252
 253 static int tun_napi_poll(struct napi_struct *napi, int budget)
 254 {
 255         unsigned int received;
 256
 257         received = tun_napi_receive(napi, budget);
 258
 259         if (received < budget)
 260                 napi_complete_done(napi, received);
 261
 262         return received;
 263 }
 264
 265 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 266                           bool napi_en, bool napi_frags)
 267 {
 268         tfile->napi_enabled = napi_en;
 269         tfile->napi_frags_enabled = napi_en && napi_frags;
 270         if (napi_en) {
 271                 netif_napi_add_tx(tun->dev, &tfile->napi, tun_napi_poll);
 272                 napi_enable(&tfile->napi);
 273         }
 274 }
 275
 276 static void tun_napi_enable(struct tun_file *tfile)
 277 {
 278         if (tfile->napi_enabled)
 279                 napi_enable(&tfile->napi);
 280 }
 281
 282 static void tun_napi_disable(struct tun_file *tfile)
 283 {
 284         if (tfile->napi_enabled)
 285                 napi_disable(&tfile->napi);
 286 }
 287
 288 static void tun_napi_del(struct tun_file *tfile)
 289 {
 290         if (tfile->napi_enabled)
 291                 netif_napi_del(&tfile->napi);
 292 }
 293
 294 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 295 {
 296         return tfile->napi_frags_enabled;
 297 }
 298
 299 #ifdef CONFIG_TUN_VNET_CROSS_LE
 300 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 301 {
 302         return tun->flags & TUN_VNET_BE ? false :
 303                 virtio_legacy_is_little_endian();
 304 }
 305
 306 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 307 {
 308         int be = !!(tun->flags & TUN_VNET_BE);
 309
 310         if (put_user(be, argp))
 311                 return -EFAULT;
 312
 313         return 0;
 314 }
 315
 316 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 317 {
 318         int be;
 319
 320         if (get_user(be, argp))
 321                 return -EFAULT;
 322
 323         if (be)
 324                 tun->flags |= TUN_VNET_BE;
 325         else
 326                 tun->flags &= ~TUN_VNET_BE;
 327
 328         return 0;
 329 }
 330 #else
 331 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 332 {
 333         return virtio_legacy_is_little_endian();
 334 }
 335
 336 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 337 {
 338         return -EINVAL;
 339 }
 340
 341 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 342 {
 343         return -EINVAL;
 344 }
 345 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 346
 347 static inline bool tun_is_little_endian(struct tun_struct *tun)
 348 {
 349         return tun->flags & TUN_VNET_LE ||
 350                 tun_legacy_is_little_endian(tun);
 351 }
 352
 353 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 354 {
 355         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 356 }
 357
 358 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 359 {
 360         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 361 }
 362
 363 static inline u32 tun_hashfn(u32 rxhash)
 364 {
 365         return rxhash & TUN_MASK_FLOW_ENTRIES;
 366 }
 367
 368 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 369 {
 370         struct tun_flow_entry *e;
 371
 372         hlist_for_each_entry_rcu(e, head, hash_link) {
 373                 if (e->rxhash == rxhash)
 374                         return e;
 375         }
 376         return NULL;
 377 }
 378
 379 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 380                                               struct hlist_head *head,
 381                                               u32 rxhash, u16 queue_index)
 382 {
 383         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 384
 385         if (e) {
 386                 netif_info(tun, tx_queued, tun->dev,
 387                            "create flow: hash %u index %u\n",
 388                            rxhash, queue_index);
 389                 e->updated = jiffies;
 390                 e->rxhash = rxhash;
 391                 e->rps_rxhash = 0;
 392                 e->queue_index = queue_index;
 393                 e->tun = tun;
 394                 hlist_add_head_rcu(&e->hash_link, head);
 395                 ++tun->flow_count;
 396         }
 397         return e;
 398 }
 399
 400 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 401 {
 402         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 403                    e->rxhash, e->queue_index);
 404         hlist_del_rcu(&e->hash_link);
 405         kfree_rcu(e, rcu);
 406         --tun->flow_count;
 407 }
 408
 409 static void tun_flow_flush(struct tun_struct *tun)
 410 {
 411         int i;
 412
 413         spin_lock_bh(&tun->lock);
 414         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 415                 struct tun_flow_entry *e;
 416                 struct hlist_node *n;
 417
 418                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 419                         tun_flow_delete(tun, e);
 420         }
 421         spin_unlock_bh(&tun->lock);
 422 }
 423
 424 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 425 {
 426         int i;
 427
 428         spin_lock_bh(&tun->lock);
 429         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 430                 struct tun_flow_entry *e;
 431                 struct hlist_node *n;
 432
 433                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 434                         if (e->queue_index == queue_index)
 435                                 tun_flow_delete(tun, e);
 436                 }
 437         }
 438         spin_unlock_bh(&tun->lock);
 439 }
 440
 441 static void tun_flow_cleanup(struct timer_list *t)
 442 {
 443         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 444         unsigned long delay = tun->ageing_time;
 445         unsigned long next_timer = jiffies + delay;
 446         unsigned long count = 0;
 447         int i;
 448
 449         spin_lock(&tun->lock);
 450         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 451                 struct tun_flow_entry *e;
 452                 struct hlist_node *n;
 453
 454                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 455                         unsigned long this_timer;
 456
 457                         this_timer = e->updated + delay;
 458                         if (time_before_eq(this_timer, jiffies)) {
 459                                 tun_flow_delete(tun, e);
 460                                 continue;
 461                         }
 462                         count++;
 463                         if (time_before(this_timer, next_timer))
 464                                 next_timer = this_timer;
 465                 }
 466         }
 467
 468         if (count)
 469                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 470         spin_unlock(&tun->lock);
 471 }
 472
 473 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 474                             struct tun_file *tfile)
 475 {
 476         struct hlist_head *head;
 477         struct tun_flow_entry *e;
 478         unsigned long delay = tun->ageing_time;
 479         u16 queue_index = tfile->queue_index;
 480
 481         head = &tun->flows[tun_hashfn(rxhash)];
 482
 483         rcu_read_lock();
 484
 485         e = tun_flow_find(head, rxhash);
 486         if (likely(e)) {
 487                 /* TODO: keep queueing to old queue until it's empty? */
 488                 if (READ_ONCE(e->queue_index) != queue_index)
 489                         WRITE_ONCE(e->queue_index, queue_index);
 490                 if (e->updated != jiffies)
 491                         e->updated = jiffies;
 492                 sock_rps_record_flow_hash(e->rps_rxhash);
 493         } else {
 494                 spin_lock_bh(&tun->lock);
 495                 if (!tun_flow_find(head, rxhash) &&
 496                     tun->flow_count < MAX_TAP_FLOWS)
 497                         tun_flow_create(tun, head, rxhash, queue_index);
 498
 499                 if (!timer_pending(&tun->flow_gc_timer))
 500                         mod_timer(&tun->flow_gc_timer,
 501                                   round_jiffies_up(jiffies + delay));
 502                 spin_unlock_bh(&tun->lock);
 503         }
 504
 505         rcu_read_unlock();
 506 }
 507
 508 /* Save the hash received in the stack receive path and update the
 509  * flow_hash table accordingly.
 510  */
 511 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 512 {
 513         if (unlikely(e->rps_rxhash != hash))
 514                 e->rps_rxhash = hash;
 515 }
 516
 517 /* We try to identify a flow through its rxhash. The reason that
 518  * we do not check rxq no. is because some cards(e.g 82599), chooses
 519  * the rxq based on the txq where the last packet of the flow comes. As
 520  * the userspace application move between processors, we may get a
 521  * different rxq no. here.
 522  */
 523 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 524 {
 525         struct tun_flow_entry *e;
 526         u32 txq = 0;
 527         u32 numqueues = 0;
 528
 529         numqueues = READ_ONCE(tun->numqueues);
 530
 531         txq = __skb_get_hash_symmetric(skb);
 532         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 533         if (e) {
 534                 tun_flow_save_rps_rxhash(e, txq);
 535                 txq = e->queue_index;
 536         } else {
 537                 /* use multiply and shift instead of expensive divide */
 538                 txq = ((u64)txq * numqueues) >> 32;
 539         }
 540
 541         return txq;
 542 }
 543
 544 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 545 {
 546         struct tun_prog *prog;
 547         u32 numqueues;
 548         u16 ret = 0;
 549
 550         numqueues = READ_ONCE(tun->numqueues);
 551         if (!numqueues)
 552                 return 0;
 553
 554         prog = rcu_dereference(tun->steering_prog);
 555         if (prog)
 556                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 557
 558         return ret % numqueues;
 559 }
 560
 561 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 562                             struct net_device *sb_dev)
 563 {
 564         struct tun_struct *tun = netdev_priv(dev);
 565         u16 ret;
 566
 567         rcu_read_lock();
 568         if (rcu_dereference(tun->steering_prog))
 569                 ret = tun_ebpf_select_queue(tun, skb);
 570         else
 571                 ret = tun_automq_select_queue(tun, skb);
 572         rcu_read_unlock();
 573
 574         return ret;
 575 }
 576
 577 static inline bool tun_not_capable(struct tun_struct *tun)
 578 {
 579         const struct cred *cred = current_cred();
 580         struct net *net = dev_net(tun->dev);
 581
 582         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 583                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 584                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 585 }
 586
 587 static void tun_set_real_num_queues(struct tun_struct *tun)
 588 {
 589         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 590         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 591 }
 592
 593 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 594 {
 595         tfile->detached = tun;
 596         list_add_tail(&tfile->next, &tun->disabled);
 597         ++tun->numdisabled;
 598 }
 599
 600 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 601 {
 602         struct tun_struct *tun = tfile->detached;
 603
 604         tfile->detached = NULL;
 605         list_del_init(&tfile->next);
 606         --tun->numdisabled;
 607         return tun;
 608 }
 609
 610 void tun_ptr_free(void *ptr)
 611 {
 612         if (!ptr)
 613                 return;
 614         if (tun_is_xdp_frame(ptr)) {
 615                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 616
 617                 xdp_return_frame(xdpf);
 618         } else {
 619                 __skb_array_destroy_skb(ptr);
 620         }
 621 }
 622 EXPORT_SYMBOL_GPL(tun_ptr_free);
 623
 624 static void tun_queue_purge(struct tun_file *tfile)
 625 {
 626         void *ptr;
 627
 628         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 629                 tun_ptr_free(ptr);
 630
 631         skb_queue_purge(&tfile->sk.sk_write_queue);
 632         skb_queue_purge(&tfile->sk.sk_error_queue);
 633 }
 634
 635 static void __tun_detach(struct tun_file *tfile, bool clean)
 636 {
 637         struct tun_file *ntfile;
 638         struct tun_struct *tun;
 639
 640         tun = rtnl_dereference(tfile->tun);
 641
 642         if (tun && clean) {
 643                 if (!tfile->detached)
 644                         tun_napi_disable(tfile);
 645                 tun_napi_del(tfile);
 646         }
 647
 648         if (tun && !tfile->detached) {
 649                 u16 index = tfile->queue_index;
 650                 BUG_ON(index >= tun->numqueues);
 651
 652                 rcu_assign_pointer(tun->tfiles[index],
 653                                    tun->tfiles[tun->numqueues - 1]);
 654                 ntfile = rtnl_dereference(tun->tfiles[index]);
 655                 ntfile->queue_index = index;
 656                 ntfile->xdp_rxq.queue_index = index;
 657                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 658                                    NULL);
 659
 660                 --tun->numqueues;
 661                 if (clean) {
 662                         RCU_INIT_POINTER(tfile->tun, NULL);
 663                         sock_put(&tfile->sk);
 664                 } else {
 665                         tun_disable_queue(tun, tfile);
 666                         tun_napi_disable(tfile);
 667                 }
 668
 669                 synchronize_net();
 670                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 671                 /* Drop read queue */
 672                 tun_queue_purge(tfile);
 673                 tun_set_real_num_queues(tun);
 674         } else if (tfile->detached && clean) {
 675                 tun = tun_enable_queue(tfile);
 676                 sock_put(&tfile->sk);
 677         }
 678
 679         if (clean) {
 680                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 681                         netif_carrier_off(tun->dev);
 682
 683                         if (!(tun->flags & IFF_PERSIST) &&
 684                             tun->dev->reg_state == NETREG_REGISTERED)
 685                                 unregister_netdevice(tun->dev);
 686                 }
 687                 if (tun)
 688                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 689                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 690         }
 691 }
 692
 693 static void tun_detach(struct tun_file *tfile, bool clean)
 694 {
 695         struct tun_struct *tun;
 696         struct net_device *dev;
 697
 698         rtnl_lock();
 699         tun = rtnl_dereference(tfile->tun);
 700         dev = tun ? tun->dev : NULL;
 701         __tun_detach(tfile, clean);
 702         if (dev)
 703                 netdev_state_change(dev);
 704         rtnl_unlock();
 705
 706         if (clean)
 707                 sock_put(&tfile->sk);
 708 }
 709
 710 static void tun_detach_all(struct net_device *dev)
 711 {
 712         struct tun_struct *tun = netdev_priv(dev);
 713         struct tun_file *tfile, *tmp;
 714         int i, n = tun->numqueues;
 715
 716         for (i = 0; i < n; i++) {
 717                 tfile = rtnl_dereference(tun->tfiles[i]);
 718                 BUG_ON(!tfile);
 719                 tun_napi_disable(tfile);
 720                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 721                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 722                 RCU_INIT_POINTER(tfile->tun, NULL);
 723                 --tun->numqueues;
 724         }
 725         list_for_each_entry(tfile, &tun->disabled, next) {
 726                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 727                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 728                 RCU_INIT_POINTER(tfile->tun, NULL);
 729         }
 730         BUG_ON(tun->numqueues != 0);
 731
 732         synchronize_net();
 733         for (i = 0; i < n; i++) {
 734                 tfile = rtnl_dereference(tun->tfiles[i]);
 735                 tun_napi_del(tfile);
 736                 /* Drop read queue */
 737                 tun_queue_purge(tfile);
 738                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 739                 sock_put(&tfile->sk);
 740         }
 741         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 742                 tun_napi_del(tfile);
 743                 tun_enable_queue(tfile);
 744                 tun_queue_purge(tfile);
 745                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 746                 sock_put(&tfile->sk);
 747         }
 748         BUG_ON(tun->numdisabled != 0);
 749
 750         if (tun->flags & IFF_PERSIST)
 751                 module_put(THIS_MODULE);
 752 }
 753
 754 static int tun_attach(struct tun_struct *tun, struct file *file,
 755                       bool skip_filter, bool napi, bool napi_frags,
 756                       bool publish_tun)
 757 {
 758         struct tun_file *tfile = file->private_data;
 759         struct net_device *dev = tun->dev;
 760         int err;
 761
 762         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 763         if (err < 0)
 764                 goto out;
 765
 766         err = -EINVAL;
 767         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 768                 goto out;
 769
 770         err = -EBUSY;
 771         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 772                 goto out;
 773
 774         err = -E2BIG;
 775         if (!tfile->detached &&
 776             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 777                 goto out;
 778
 779         err = 0;
 780
 781         /* Re-attach the filter to persist device */
 782         if (!skip_filter && (tun->filter_attached == true)) {
 783                 lock_sock(tfile->socket.sk);
 784                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 785                 release_sock(tfile->socket.sk);
 786                 if (!err)
 787                         goto out;
 788         }
 789
 790         if (!tfile->detached &&
 791             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 792                             GFP_KERNEL, tun_ptr_free)) {
 793                 err = -ENOMEM;
 794                 goto out;
 795         }
 796
 797         tfile->queue_index = tun->numqueues;
 798         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 799
 800         if (tfile->detached) {
 801                 /* Re-attach detached tfile, updating XDP queue_index */
 802                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 803
 804                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 805                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 806         } else {
 807                 /* Setup XDP RX-queue info, for new tfile getting attached */
 808                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 809                                        tun->dev, tfile->queue_index, 0);
 810                 if (err < 0)
 811                         goto out;
 812                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 813                                                  MEM_TYPE_PAGE_SHARED, NULL);
 814                 if (err < 0) {
 815                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 816                         goto out;
 817                 }
 818                 err = 0;
 819         }
 820
 821         if (tfile->detached) {
 822                 tun_enable_queue(tfile);
 823                 tun_napi_enable(tfile);
 824         } else {
 825                 sock_hold(&tfile->sk);
 826                 tun_napi_init(tun, tfile, napi, napi_frags);
 827         }
 828
 829         if (rtnl_dereference(tun->xdp_prog))
 830                 sock_set_flag(&tfile->sk, SOCK_XDP);
 831
 832         /* device is allowed to go away first, so no need to hold extra
 833          * refcnt.
 834          */
 835
 836         /* Publish tfile->tun and tun->tfiles only after we've fully
 837          * initialized tfile; otherwise we risk using half-initialized
 838          * object.
 839          */
 840         if (publish_tun)
 841                 rcu_assign_pointer(tfile->tun, tun);
 842         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 843         tun->numqueues++;
 844         tun_set_real_num_queues(tun);
 845 out:
 846         return err;
 847 }
 848
 849 static struct tun_struct *tun_get(struct tun_file *tfile)
 850 {
 851         struct tun_struct *tun;
 852
 853         rcu_read_lock();
 854         tun = rcu_dereference(tfile->tun);
 855         if (tun)
 856                 dev_hold(tun->dev);
 857         rcu_read_unlock();
 858
 859         return tun;
 860 }
 861
 862 static void tun_put(struct tun_struct *tun)
 863 {
 864         dev_put(tun->dev);
 865 }
 866
 867 /* TAP filtering */
 868 static void addr_hash_set(u32 *mask, const u8 *addr)
 869 {
 870         int n = ether_crc(ETH_ALEN, addr) >> 26;
 871         mask[n >> 5] |= (1 << (n & 31));
 872 }
 873
 874 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 875 {
 876         int n = ether_crc(ETH_ALEN, addr) >> 26;
 877         return mask[n >> 5] & (1 << (n & 31));
 878 }
 879
 880 static int update_filter(struct tap_filter *filter, void __user *arg)
 881 {
 882         struct { u8 u[ETH_ALEN]; } *addr;
 883         struct tun_filter uf;
 884         int err, alen, n, nexact;
 885
 886         if (copy_from_user(&uf, arg, sizeof(uf)))
 887                 return -EFAULT;
 888
 889         if (!uf.count) {
 890                 /* Disabled */
 891                 filter->count = 0;
 892                 return 0;
 893         }
 894
 895         alen = ETH_ALEN * uf.count;
 896         addr = memdup_user(arg + sizeof(uf), alen);
 897         if (IS_ERR(addr))
 898                 return PTR_ERR(addr);
 899
 900         /* The filter is updated without holding any locks. Which is
 901          * perfectly safe. We disable it first and in the worst
 902          * case we'll accept a few undesired packets. */
 903         filter->count = 0;
 904         wmb();
 905
 906         /* Use first set of addresses as an exact filter */
 907         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 908                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 909
 910         nexact = n;
 911
 912         /* Remaining multicast addresses are hashed,
 913          * unicast will leave the filter disabled. */
 914         memset(filter->mask, 0, sizeof(filter->mask));
 915         for (; n < uf.count; n++) {
 916                 if (!is_multicast_ether_addr(addr[n].u)) {
 917                         err = 0; /* no filter */
 918                         goto free_addr;
 919                 }
 920                 addr_hash_set(filter->mask, addr[n].u);
 921         }
 922
 923         /* For ALLMULTI just set the mask to all ones.
 924          * This overrides the mask populated above. */
 925         if ((uf.flags & TUN_FLT_ALLMULTI))
 926                 memset(filter->mask, ~0, sizeof(filter->mask));
 927
 928         /* Now enable the filter */
 929         wmb();
 930         filter->count = nexact;
 931
 932         /* Return the number of exact filters */
 933         err = nexact;
 934 free_addr:
 935         kfree(addr);
 936         return err;
 937 }
 938
 939 /* Returns: 0 - drop, !=0 - accept */
 940 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 941 {
 942         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 943          * at this point. */
 944         struct ethhdr *eh = (struct ethhdr *) skb->data;
 945         int i;
 946
 947         /* Exact match */
 948         for (i = 0; i < filter->count; i++)
 949                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 950                         return 1;
 951
 952         /* Inexact match (multicast only) */
 953         if (is_multicast_ether_addr(eh->h_dest))
 954                 return addr_hash_test(filter->mask, eh->h_dest);
 955
 956         return 0;
 957 }
 958
 959 /*
 960  * Checks whether the packet is accepted or not.
 961  * Returns: 0 - drop, !=0 - accept
 962  */
 963 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 964 {
 965         if (!filter->count)
 966                 return 1;
 967
 968         return run_filter(filter, skb);
 969 }
 970
 971 /* Network device part of the driver */
 972
 973 static const struct ethtool_ops tun_ethtool_ops;
 974
 975 static int tun_net_init(struct net_device *dev)
 976 {
 977         struct tun_struct *tun = netdev_priv(dev);
 978         struct ifreq *ifr = tun->ifr;
 979         int err;
 980
 981         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 982         if (!dev->tstats)
 983                 return -ENOMEM;
 984
 985         spin_lock_init(&tun->lock);
 986
 987         err = security_tun_dev_alloc_security(&tun->security);
 988         if (err < 0) {
 989                 free_percpu(dev->tstats);
 990                 return err;
 991         }
 992
 993         tun_flow_init(tun);
 994
 995         dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
 996                            TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
 997                            NETIF_F_HW_VLAN_STAG_TX;
 998         dev->features = dev->hw_features | NETIF_F_LLTX;
 999         dev->vlan_features = dev->features &
1000                              ~(NETIF_F_HW_VLAN_CTAG_TX |
1001                                NETIF_F_HW_VLAN_STAG_TX);
1002
1003         tun->flags = (tun->flags & ~TUN_FEATURES) |
1004                       (ifr->ifr_flags & TUN_FEATURES);
1005
1006         INIT_LIST_HEAD(&tun->disabled);
1007         err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1008                          ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1009         if (err < 0) {
1010                 tun_flow_uninit(tun);
1011                 security_tun_dev_free_security(tun->security);
1012                 free_percpu(dev->tstats);
1013                 return err;
1014         }
1015         return 0;
1016 }
1017
1018 /* Net device detach from fd. */
1019 static void tun_net_uninit(struct net_device *dev)
1020 {
1021         tun_detach_all(dev);
1022 }
1023
1024 /* Net device open. */
1025 static int tun_net_open(struct net_device *dev)
1026 {
1027         netif_tx_start_all_queues(dev);
1028
1029         return 0;
1030 }
1031
1032 /* Net device close. */
1033 static int tun_net_close(struct net_device *dev)
1034 {
1035         netif_tx_stop_all_queues(dev);
1036         return 0;
1037 }
1038
1039 /* Net device start xmit */
1040 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1041 {
1042 #ifdef CONFIG_RPS
1043         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1044                 /* Select queue was not called for the skbuff, so we extract the
1045                  * RPS hash and save it into the flow_table here.
1046                  */
1047                 struct tun_flow_entry *e;
1048                 __u32 rxhash;
1049
1050                 rxhash = __skb_get_hash_symmetric(skb);
1051                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1052                 if (e)
1053                         tun_flow_save_rps_rxhash(e, rxhash);
1054         }
1055 #endif
1056 }
1057
1058 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1059                                     struct sk_buff *skb,
1060                                     int len)
1061 {
1062         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1063
1064         if (prog)
1065                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1066
1067         return len;
1068 }
1069
1070 /* Net device start xmit */
1071 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1072 {
1073         struct tun_struct *tun = netdev_priv(dev);
1074         enum skb_drop_reason drop_reason;
1075         int txq = skb->queue_mapping;
1076         struct netdev_queue *queue;
1077         struct tun_file *tfile;
1078         int len = skb->len;
1079
1080         rcu_read_lock();
1081         tfile = rcu_dereference(tun->tfiles[txq]);
1082
1083         /* Drop packet if interface is not attached */
1084         if (!tfile) {
1085                 drop_reason = SKB_DROP_REASON_DEV_READY;
1086                 goto drop;
1087         }
1088
1089         if (!rcu_dereference(tun->steering_prog))
1090                 tun_automq_xmit(tun, skb);
1091
1092         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1093
1094         /* Drop if the filter does not like it.
1095          * This is a noop if the filter is disabled.
1096          * Filter can be enabled only for the TAP devices. */
1097         if (!check_filter(&tun->txflt, skb)) {
1098                 drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1099                 goto drop;
1100         }
1101
1102         if (tfile->socket.sk->sk_filter &&
1103             sk_filter(tfile->socket.sk, skb)) {
1104                 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1105                 goto drop;
1106         }
1107
1108         len = run_ebpf_filter(tun, skb, len);
1109         if (len == 0) {
1110                 drop_reason = SKB_DROP_REASON_TAP_FILTER;
1111                 goto drop;
1112         }
1113
1114         if (pskb_trim(skb, len)) {
1115                 drop_reason = SKB_DROP_REASON_NOMEM;
1116                 goto drop;
1117         }
1118
1119         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1120                 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1121                 goto drop;
1122         }
1123
1124         skb_tx_timestamp(skb);
1125
1126         /* Orphan the skb - required as we might hang on to it
1127          * for indefinite time.
1128          */
1129         skb_orphan(skb);
1130
1131         nf_reset_ct(skb);
1132
1133         if (ptr_ring_produce(&tfile->tx_ring, skb)) {
1134                 drop_reason = SKB_DROP_REASON_FULL_RING;
1135                 goto drop;
1136         }
1137
1138         /* NETIF_F_LLTX requires to do our own update of trans_start */
1139         queue = netdev_get_tx_queue(dev, txq);
1140         txq_trans_cond_update(queue);
1141
1142         /* Notify and wake up reader process */
1143         if (tfile->flags & TUN_FASYNC)
1144                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1145         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1146
1147         rcu_read_unlock();
1148         return NETDEV_TX_OK;
1149
1150 drop:
1151         dev_core_stats_tx_dropped_inc(dev);
1152         skb_tx_error(skb);
1153         kfree_skb_reason(skb, drop_reason);
1154         rcu_read_unlock();
1155         return NET_XMIT_DROP;
1156 }
1157
1158 static void tun_net_mclist(struct net_device *dev)
1159 {
1160         /*
1161          * This callback is supposed to deal with mc filter in
1162          * _rx_ path and has nothing to do with the _tx_ path.
1163          * In rx path we always accept everything userspace gives us.
1164          */
1165 }
1166
1167 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1168         netdev_features_t features)
1169 {
1170         struct tun_struct *tun = netdev_priv(dev);
1171
1172         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1173 }
1174
1175 static void tun_set_headroom(struct net_device *dev, int new_hr)
1176 {
1177         struct tun_struct *tun = netdev_priv(dev);
1178
1179         if (new_hr < NET_SKB_PAD)
1180                 new_hr = NET_SKB_PAD;
1181
1182         tun->align = new_hr;
1183 }
1184
1185 static void
1186 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1187 {
1188         struct tun_struct *tun = netdev_priv(dev);
1189
1190         dev_get_tstats64(dev, stats);
1191
1192         stats->rx_frame_errors +=
1193                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1194 }
1195
1196 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1197                        struct netlink_ext_ack *extack)
1198 {
1199         struct tun_struct *tun = netdev_priv(dev);
1200         struct tun_file *tfile;
1201         struct bpf_prog *old_prog;
1202         int i;
1203
1204         old_prog = rtnl_dereference(tun->xdp_prog);
1205         rcu_assign_pointer(tun->xdp_prog, prog);
1206         if (old_prog)
1207                 bpf_prog_put(old_prog);
1208
1209         for (i = 0; i < tun->numqueues; i++) {
1210                 tfile = rtnl_dereference(tun->tfiles[i]);
1211                 if (prog)
1212                         sock_set_flag(&tfile->sk, SOCK_XDP);
1213                 else
1214                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1215         }
1216         list_for_each_entry(tfile, &tun->disabled, next) {
1217                 if (prog)
1218                         sock_set_flag(&tfile->sk, SOCK_XDP);
1219                 else
1220                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1221         }
1222
1223         return 0;
1224 }
1225
1226 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1227 {
1228         switch (xdp->command) {
1229         case XDP_SETUP_PROG:
1230                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1231         default:
1232                 return -EINVAL;
1233         }
1234 }
1235
1236 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1237 {
1238         if (new_carrier) {
1239                 struct tun_struct *tun = netdev_priv(dev);
1240
1241                 if (!tun->numqueues)
1242                         return -EPERM;
1243
1244                 netif_carrier_on(dev);
1245         } else {
1246                 netif_carrier_off(dev);
1247         }
1248         return 0;
1249 }
1250
1251 static const struct net_device_ops tun_netdev_ops = {
1252         .ndo_init               = tun_net_init,
1253         .ndo_uninit             = tun_net_uninit,
1254         .ndo_open               = tun_net_open,
1255         .ndo_stop               = tun_net_close,
1256         .ndo_start_xmit         = tun_net_xmit,
1257         .ndo_fix_features       = tun_net_fix_features,
1258         .ndo_select_queue       = tun_select_queue,
1259         .ndo_set_rx_headroom    = tun_set_headroom,
1260         .ndo_get_stats64        = tun_net_get_stats64,
1261         .ndo_change_carrier     = tun_net_change_carrier,
1262 };
1263
1264 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1265 {
1266         /* Notify and wake up reader process */
1267         if (tfile->flags & TUN_FASYNC)
1268                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1269         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1270 }
1271
1272 static int tun_xdp_xmit(struct net_device *dev, int n,
1273                         struct xdp_frame **frames, u32 flags)
1274 {
1275         struct tun_struct *tun = netdev_priv(dev);
1276         struct tun_file *tfile;
1277         u32 numqueues;
1278         int nxmit = 0;
1279         int i;
1280
1281         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1282                 return -EINVAL;
1283
1284         rcu_read_lock();
1285
1286 resample:
1287         numqueues = READ_ONCE(tun->numqueues);
1288         if (!numqueues) {
1289                 rcu_read_unlock();
1290                 return -ENXIO; /* Caller will free/return all frames */
1291         }
1292
1293         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1294                                             numqueues]);
1295         if (unlikely(!tfile))
1296                 goto resample;
1297
1298         spin_lock(&tfile->tx_ring.producer_lock);
1299         for (i = 0; i < n; i++) {
1300                 struct xdp_frame *xdp = frames[i];
1301                 /* Encode the XDP flag into lowest bit for consumer to differ
1302                  * XDP buffer from sk_buff.
1303                  */
1304                 void *frame = tun_xdp_to_ptr(xdp);
1305
1306                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1307                         dev_core_stats_tx_dropped_inc(dev);
1308                         break;
1309                 }
1310                 nxmit++;
1311         }
1312         spin_unlock(&tfile->tx_ring.producer_lock);
1313
1314         if (flags & XDP_XMIT_FLUSH)
1315                 __tun_xdp_flush_tfile(tfile);
1316
1317         rcu_read_unlock();
1318         return nxmit;
1319 }
1320
1321 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1322 {
1323         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1324         int nxmit;
1325
1326         if (unlikely(!frame))
1327                 return -EOVERFLOW;
1328
1329         nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1330         if (!nxmit)
1331                 xdp_return_frame_rx_napi(frame);
1332         return nxmit;
1333 }
1334
1335 static const struct net_device_ops tap_netdev_ops = {
1336         .ndo_init               = tun_net_init,
1337         .ndo_uninit             = tun_net_uninit,
1338         .ndo_open               = tun_net_open,
1339         .ndo_stop               = tun_net_close,
1340         .ndo_start_xmit         = tun_net_xmit,
1341         .ndo_fix_features       = tun_net_fix_features,
1342         .ndo_set_rx_mode        = tun_net_mclist,
1343         .ndo_set_mac_address    = eth_mac_addr,
1344         .ndo_validate_addr      = eth_validate_addr,
1345         .ndo_select_queue       = tun_select_queue,
1346         .ndo_features_check     = passthru_features_check,
1347         .ndo_set_rx_headroom    = tun_set_headroom,
1348         .ndo_get_stats64        = dev_get_tstats64,
1349         .ndo_bpf                = tun_xdp,
1350         .ndo_xdp_xmit           = tun_xdp_xmit,
1351         .ndo_change_carrier     = tun_net_change_carrier,
1352 };
1353
1354 static void tun_flow_init(struct tun_struct *tun)
1355 {
1356         int i;
1357
1358         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1359                 INIT_HLIST_HEAD(&tun->flows[i]);
1360
1361         tun->ageing_time = TUN_FLOW_EXPIRE;
1362         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1363         mod_timer(&tun->flow_gc_timer,
1364                   round_jiffies_up(jiffies + tun->ageing_time));
1365 }
1366
1367 static void tun_flow_uninit(struct tun_struct *tun)
1368 {
1369         del_timer_sync(&tun->flow_gc_timer);
1370         tun_flow_flush(tun);
1371 }
1372
1373 #define MIN_MTU 68
1374 #define MAX_MTU 65535
1375
1376 /* Initialize net device. */
1377 static void tun_net_initialize(struct net_device *dev)
1378 {
1379         struct tun_struct *tun = netdev_priv(dev);
1380
1381         switch (tun->flags & TUN_TYPE_MASK) {
1382         case IFF_TUN:
1383                 dev->netdev_ops = &tun_netdev_ops;
1384                 dev->header_ops = &ip_tunnel_header_ops;
1385
1386                 /* Point-to-Point TUN Device */
1387                 dev->hard_header_len = 0;
1388                 dev->addr_len = 0;
1389                 dev->mtu = 1500;
1390
1391                 /* Zero header length */
1392                 dev->type = ARPHRD_NONE;
1393                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1394                 break;
1395
1396         case IFF_TAP:
1397                 dev->netdev_ops = &tap_netdev_ops;
1398                 /* Ethernet TAP Device */
1399                 ether_setup(dev);
1400                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1401                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1402
1403                 eth_hw_addr_random(dev);
1404
1405                 /* Currently tun does not support XDP, only tap does. */
1406                 dev->xdp_features = NETDEV_XDP_ACT_BASIC |
1407                                     NETDEV_XDP_ACT_REDIRECT |
1408                                     NETDEV_XDP_ACT_NDO_XMIT;
1409
1410                 break;
1411         }
1412
1413         dev->min_mtu = MIN_MTU;
1414         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1415 }
1416
1417 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1418 {
1419         struct sock *sk = tfile->socket.sk;
1420
1421         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1422 }
1423
1424 /* Character device part */
1425
1426 /* Poll */
1427 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1428 {
1429         struct tun_file *tfile = file->private_data;
1430         struct tun_struct *tun = tun_get(tfile);
1431         struct sock *sk;
1432         __poll_t mask = 0;
1433
1434         if (!tun)
1435                 return EPOLLERR;
1436
1437         sk = tfile->socket.sk;
1438
1439         poll_wait(file, sk_sleep(sk), wait);
1440
1441         if (!ptr_ring_empty(&tfile->tx_ring))
1442                 mask |= EPOLLIN | EPOLLRDNORM;
1443
1444         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1445          * guarantee EPOLLOUT to be raised by either here or
1446          * tun_sock_write_space(). Then process could get notification
1447          * after it writes to a down device and meets -EIO.
1448          */
1449         if (tun_sock_writeable(tun, tfile) ||
1450             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1451              tun_sock_writeable(tun, tfile)))
1452                 mask |= EPOLLOUT | EPOLLWRNORM;
1453
1454         if (tun->dev->reg_state != NETREG_REGISTERED)
1455                 mask = EPOLLERR;
1456
1457         tun_put(tun);
1458         return mask;
1459 }
1460
1461 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1462                                             size_t len,
1463                                             const struct iov_iter *it)
1464 {
1465         struct sk_buff *skb;
1466         size_t linear;
1467         int err;
1468         int i;
1469
1470         if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1471             len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1472                 return ERR_PTR(-EMSGSIZE);
1473
1474         local_bh_disable();
1475         skb = napi_get_frags(&tfile->napi);
1476         local_bh_enable();
1477         if (!skb)
1478                 return ERR_PTR(-ENOMEM);
1479
1480         linear = iov_iter_single_seg_count(it);
1481         err = __skb_grow(skb, linear);
1482         if (err)
1483                 goto free;
1484
1485         skb->len = len;
1486         skb->data_len = len - linear;
1487         skb->truesize += skb->data_len;
1488
1489         for (i = 1; i < it->nr_segs; i++) {
1490                 const struct iovec *iov = iter_iov(it);
1491                 size_t fragsz = iov->iov_len;
1492                 struct page *page;
1493                 void *frag;
1494
1495                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1496                         err = -EINVAL;
1497                         goto free;
1498                 }
1499                 frag = netdev_alloc_frag(fragsz);
1500                 if (!frag) {
1501                         err = -ENOMEM;
1502                         goto free;
1503                 }
1504                 page = virt_to_head_page(frag);
1505                 skb_fill_page_desc(skb, i - 1, page,
1506                                    frag - page_address(page), fragsz);
1507         }
1508
1509         return skb;
1510 free:
1511         /* frees skb and all frags allocated with napi_alloc_frag() */
1512         napi_free_frags(&tfile->napi);
1513         return ERR_PTR(err);
1514 }
1515
1516 /* prepad is the amount to reserve at front.  len is length after that.
1517  * linear is a hint as to how much to copy (usually headers). */
1518 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1519                                      size_t prepad, size_t len,
1520                                      size_t linear, int noblock)
1521 {
1522         struct sock *sk = tfile->socket.sk;
1523         struct sk_buff *skb;
1524         int err;
1525
1526         /* Under a page?  Don't bother with paged skb. */
1527         if (prepad + len < PAGE_SIZE)
1528                 linear = len;
1529
1530         if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
1531                 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
1532         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1533                                    &err, PAGE_ALLOC_COSTLY_ORDER);
1534         if (!skb)
1535                 return ERR_PTR(err);
1536
1537         skb_reserve(skb, prepad);
1538         skb_put(skb, linear);
1539         skb->data_len = len - linear;
1540         skb->len += len - linear;
1541
1542         return skb;
1543 }
1544
1545 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1546                            struct sk_buff *skb, int more)
1547 {
1548         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1549         struct sk_buff_head process_queue;
1550         u32 rx_batched = tun->rx_batched;
1551         bool rcv = false;
1552
1553         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1554                 local_bh_disable();
1555                 skb_record_rx_queue(skb, tfile->queue_index);
1556                 netif_receive_skb(skb);
1557                 local_bh_enable();
1558                 return;
1559         }
1560
1561         spin_lock(&queue->lock);
1562         if (!more || skb_queue_len(queue) == rx_batched) {
1563                 __skb_queue_head_init(&process_queue);
1564                 skb_queue_splice_tail_init(queue, &process_queue);
1565                 rcv = true;
1566         } else {
1567                 __skb_queue_tail(queue, skb);
1568         }
1569         spin_unlock(&queue->lock);
1570
1571         if (rcv) {
1572                 struct sk_buff *nskb;
1573
1574                 local_bh_disable();
1575                 while ((nskb = __skb_dequeue(&process_queue))) {
1576                         skb_record_rx_queue(nskb, tfile->queue_index);
1577                         netif_receive_skb(nskb);
1578                 }
1579                 skb_record_rx_queue(skb, tfile->queue_index);
1580                 netif_receive_skb(skb);
1581                 local_bh_enable();
1582         }
1583 }
1584
1585 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1586                               int len, int noblock, bool zerocopy)
1587 {
1588         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1589                 return false;
1590
1591         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1592                 return false;
1593
1594         if (!noblock)
1595                 return false;
1596
1597         if (zerocopy)
1598                 return false;
1599
1600         if (SKB_DATA_ALIGN(len + TUN_RX_PAD + XDP_PACKET_HEADROOM) +
1601             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1602                 return false;
1603
1604         return true;
1605 }
1606
1607 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1608                                        struct page_frag *alloc_frag, char *buf,
1609                                        int buflen, int len, int pad)
1610 {
1611         struct sk_buff *skb = build_skb(buf, buflen);
1612
1613         if (!skb)
1614                 return ERR_PTR(-ENOMEM);
1615
1616         skb_reserve(skb, pad);
1617         skb_put(skb, len);
1618         skb_set_owner_w(skb, tfile->socket.sk);
1619
1620         get_page(alloc_frag->page);
1621         alloc_frag->offset += buflen;
1622
1623         return skb;
1624 }
1625
1626 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1627                        struct xdp_buff *xdp, u32 act)
1628 {
1629         int err;
1630
1631         switch (act) {
1632         case XDP_REDIRECT:
1633                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1634                 if (err) {
1635                         dev_core_stats_rx_dropped_inc(tun->dev);
1636                         return err;
1637                 }
1638                 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1639                 break;
1640         case XDP_TX:
1641                 err = tun_xdp_tx(tun->dev, xdp);
1642                 if (err < 0) {
1643                         dev_core_stats_rx_dropped_inc(tun->dev);
1644                         return err;
1645                 }
1646                 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1647                 break;
1648         case XDP_PASS:
1649                 break;
1650         default:
1651                 bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1652                 fallthrough;
1653         case XDP_ABORTED:
1654                 trace_xdp_exception(tun->dev, xdp_prog, act);
1655                 fallthrough;
1656         case XDP_DROP:
1657                 dev_core_stats_rx_dropped_inc(tun->dev);
1658                 break;
1659         }
1660
1661         return act;
1662 }
1663
1664 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1665                                      struct tun_file *tfile,
1666                                      struct iov_iter *from,
1667                                      struct virtio_net_hdr *hdr,
1668                                      int len, int *skb_xdp)
1669 {
1670         struct page_frag *alloc_frag = &current->task_frag;
1671         struct bpf_prog *xdp_prog;
1672         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1673         char *buf;
1674         size_t copied;
1675         int pad = TUN_RX_PAD;
1676         int err = 0;
1677
1678         rcu_read_lock();
1679         xdp_prog = rcu_dereference(tun->xdp_prog);
1680         if (xdp_prog)
1681                 pad += XDP_PACKET_HEADROOM;
1682         buflen += SKB_DATA_ALIGN(len + pad);
1683         rcu_read_unlock();
1684
1685         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1686         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1687                 return ERR_PTR(-ENOMEM);
1688
1689         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1690         copied = copy_page_from_iter(alloc_frag->page,
1691                                      alloc_frag->offset + pad,
1692                                      len, from);
1693         if (copied != len)
1694                 return ERR_PTR(-EFAULT);
1695
1696         /* There's a small window that XDP may be set after the check
1697          * of xdp_prog above, this should be rare and for simplicity
1698          * we do XDP on skb in case the headroom is not enough.
1699          */
1700         if (hdr->gso_type || !xdp_prog) {
1701                 *skb_xdp = 1;
1702                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1703                                        pad);
1704         }
1705
1706         *skb_xdp = 0;
1707
1708         local_bh_disable();
1709         rcu_read_lock();
1710         xdp_prog = rcu_dereference(tun->xdp_prog);
1711         if (xdp_prog) {
1712                 struct xdp_buff xdp;
1713                 u32 act;
1714
1715                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1716                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1717
1718                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1719                 if (act == XDP_REDIRECT || act == XDP_TX) {
1720                         get_page(alloc_frag->page);
1721                         alloc_frag->offset += buflen;
1722                 }
1723                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1724                 if (err < 0) {
1725                         if (act == XDP_REDIRECT || act == XDP_TX)
1726                                 put_page(alloc_frag->page);
1727                         goto out;
1728                 }
1729
1730                 if (err == XDP_REDIRECT)
1731                         xdp_do_flush();
1732                 if (err != XDP_PASS)
1733                         goto out;
1734
1735                 pad = xdp.data - xdp.data_hard_start;
1736                 len = xdp.data_end - xdp.data;
1737         }
1738         rcu_read_unlock();
1739         local_bh_enable();
1740
1741         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1742
1743 out:
1744         rcu_read_unlock();
1745         local_bh_enable();
1746         return NULL;
1747 }
1748
1749 /* Get packet from user space buffer */
1750 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1751                             void *msg_control, struct iov_iter *from,
1752                             int noblock, bool more)
1753 {
1754         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1755         struct sk_buff *skb;
1756         size_t total_len = iov_iter_count(from);
1757         size_t len = total_len, align = tun->align, linear;
1758         struct virtio_net_hdr gso = { 0 };
1759         int good_linear;
1760         int copylen;
1761         bool zerocopy = false;
1762         int err;
1763         u32 rxhash = 0;
1764         int skb_xdp = 1;
1765         bool frags = tun_napi_frags_enabled(tfile);
1766         enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1767
1768         if (!(tun->flags & IFF_NO_PI)) {
1769                 if (len < sizeof(pi))
1770                         return -EINVAL;
1771                 len -= sizeof(pi);
1772
1773                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1774                         return -EFAULT;
1775         }
1776
1777         if (tun->flags & IFF_VNET_HDR) {
1778                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1779
1780                 if (len < vnet_hdr_sz)
1781                         return -EINVAL;
1782                 len -= vnet_hdr_sz;
1783
1784                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1785                         return -EFAULT;
1786
1787                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1788                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1789                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1790
1791                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1792                         return -EINVAL;
1793                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1794         }
1795
1796         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1797                 align += NET_IP_ALIGN;
1798                 if (unlikely(len < ETH_HLEN ||
1799                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1800                         return -EINVAL;
1801         }
1802
1803         good_linear = SKB_MAX_HEAD(align);
1804
1805         if (msg_control) {
1806                 struct iov_iter i = *from;
1807
1808                 /* There are 256 bytes to be copied in skb, so there is
1809                  * enough room for skb expand head in case it is used.
1810                  * The rest of the buffer is mapped from userspace.
1811                  */
1812                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1813                 if (copylen > good_linear)
1814                         copylen = good_linear;
1815                 linear = copylen;
1816                 iov_iter_advance(&i, copylen);
1817                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1818                         zerocopy = true;
1819         }
1820
1821         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1822                 /* For the packet that is not easy to be processed
1823                  * (e.g gso or jumbo packet), we will do it at after
1824                  * skb was created with generic XDP routine.
1825                  */
1826                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1827                 err = PTR_ERR_OR_ZERO(skb);
1828                 if (err)
1829                         goto drop;
1830                 if (!skb)
1831                         return total_len;
1832         } else {
1833                 if (!zerocopy) {
1834                         copylen = len;
1835                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1836                                 linear = good_linear;
1837                         else
1838                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1839                 }
1840
1841                 if (frags) {
1842                         mutex_lock(&tfile->napi_mutex);
1843                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1844                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1845                          * If zerocopy is enabled, then this layout will be
1846                          * overwritten by zerocopy_sg_from_iter().
1847                          */
1848                         zerocopy = false;
1849                 } else {
1850                         if (!linear)
1851                                 linear = min_t(size_t, good_linear, copylen);
1852
1853                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1854                                             noblock);
1855                 }
1856
1857                 err = PTR_ERR_OR_ZERO(skb);
1858                 if (err)
1859                         goto drop;
1860
1861                 if (zerocopy)
1862                         err = zerocopy_sg_from_iter(skb, from);
1863                 else
1864                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1865
1866                 if (err) {
1867                         err = -EFAULT;
1868                         drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1869                         goto drop;
1870                 }
1871         }
1872
1873         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1874                 atomic_long_inc(&tun->rx_frame_errors);
1875                 err = -EINVAL;
1876                 goto free_skb;
1877         }
1878
1879         switch (tun->flags & TUN_TYPE_MASK) {
1880         case IFF_TUN:
1881                 if (tun->flags & IFF_NO_PI) {
1882                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1883
1884                         switch (ip_version) {
1885                         case 4:
1886                                 pi.proto = htons(ETH_P_IP);
1887                                 break;
1888                         case 6:
1889                                 pi.proto = htons(ETH_P_IPV6);
1890                                 break;
1891                         default:
1892                                 err = -EINVAL;
1893                                 goto drop;
1894                         }
1895                 }
1896
1897                 skb_reset_mac_header(skb);
1898                 skb->protocol = pi.proto;
1899                 skb->dev = tun->dev;
1900                 break;
1901         case IFF_TAP:
1902                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1903                         err = -ENOMEM;
1904                         drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1905                         goto drop;
1906                 }
1907                 skb->protocol = eth_type_trans(skb, tun->dev);
1908                 break;
1909         }
1910
1911         /* copy skb_ubuf_info for callback when skb has no error */
1912         if (zerocopy) {
1913                 skb_zcopy_init(skb, msg_control);
1914         } else if (msg_control) {
1915                 struct ubuf_info *uarg = msg_control;
1916                 uarg->callback(NULL, uarg, false);
1917         }
1918
1919         skb_reset_network_header(skb);
1920         skb_probe_transport_header(skb);
1921         skb_record_rx_queue(skb, tfile->queue_index);
1922
1923         if (skb_xdp) {
1924                 struct bpf_prog *xdp_prog;
1925                 int ret;
1926
1927                 local_bh_disable();
1928                 rcu_read_lock();
1929                 xdp_prog = rcu_dereference(tun->xdp_prog);
1930                 if (xdp_prog) {
1931                         ret = do_xdp_generic(xdp_prog, skb);
1932                         if (ret != XDP_PASS) {
1933                                 rcu_read_unlock();
1934                                 local_bh_enable();
1935                                 goto unlock_frags;
1936                         }
1937                 }
1938                 rcu_read_unlock();
1939                 local_bh_enable();
1940         }
1941
1942         /* Compute the costly rx hash only if needed for flow updates.
1943          * We may get a very small possibility of OOO during switching, not
1944          * worth to optimize.
1945          */
1946         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1947             !tfile->detached)
1948                 rxhash = __skb_get_hash_symmetric(skb);
1949
1950         rcu_read_lock();
1951         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1952                 err = -EIO;
1953                 rcu_read_unlock();
1954                 drop_reason = SKB_DROP_REASON_DEV_READY;
1955                 goto drop;
1956         }
1957
1958         if (frags) {
1959                 u32 headlen;
1960
1961                 /* Exercise flow dissector code path. */
1962                 skb_push(skb, ETH_HLEN);
1963                 headlen = eth_get_headlen(tun->dev, skb->data,
1964                                           skb_headlen(skb));
1965
1966                 if (unlikely(headlen > skb_headlen(skb))) {
1967                         WARN_ON_ONCE(1);
1968                         err = -ENOMEM;
1969                         dev_core_stats_rx_dropped_inc(tun->dev);
1970 napi_busy:
1971                         napi_free_frags(&tfile->napi);
1972                         rcu_read_unlock();
1973                         mutex_unlock(&tfile->napi_mutex);
1974                         return err;
1975                 }
1976
1977                 if (likely(napi_schedule_prep(&tfile->napi))) {
1978                         local_bh_disable();
1979                         napi_gro_frags(&tfile->napi);
1980                         napi_complete(&tfile->napi);
1981                         local_bh_enable();
1982                 } else {
1983                         err = -EBUSY;
1984                         goto napi_busy;
1985                 }
1986                 mutex_unlock(&tfile->napi_mutex);
1987         } else if (tfile->napi_enabled) {
1988                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1989                 int queue_len;
1990
1991                 spin_lock_bh(&queue->lock);
1992
1993                 if (unlikely(tfile->detached)) {
1994                         spin_unlock_bh(&queue->lock);
1995                         rcu_read_unlock();
1996                         err = -EBUSY;
1997                         goto free_skb;
1998                 }
1999
2000                 __skb_queue_tail(queue, skb);
2001                 queue_len = skb_queue_len(queue);
2002                 spin_unlock(&queue->lock);
2003
2004                 if (!more || queue_len > NAPI_POLL_WEIGHT)
2005                         napi_schedule(&tfile->napi);
2006
2007                 local_bh_enable();
2008         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2009                 tun_rx_batched(tun, tfile, skb, more);
2010         } else {
2011                 netif_rx(skb);
2012         }
2013         rcu_read_unlock();
2014
2015         preempt_disable();
2016         dev_sw_netstats_rx_add(tun->dev, len);
2017         preempt_enable();
2018
2019         if (rxhash)
2020                 tun_flow_update(tun, rxhash, tfile);
2021
2022         return total_len;
2023
2024 drop:
2025         if (err != -EAGAIN)
2026                 dev_core_stats_rx_dropped_inc(tun->dev);
2027
2028 free_skb:
2029         if (!IS_ERR_OR_NULL(skb))
2030                 kfree_skb_reason(skb, drop_reason);
2031
2032 unlock_frags:
2033         if (frags) {
2034                 tfile->napi.skb = NULL;
2035                 mutex_unlock(&tfile->napi_mutex);
2036         }
2037
2038         return err ?: total_len;
2039 }
2040
2041 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2042 {
2043         struct file *file = iocb->ki_filp;
2044         struct tun_file *tfile = file->private_data;
2045         struct tun_struct *tun = tun_get(tfile);
2046         ssize_t result;
2047         int noblock = 0;
2048
2049         if (!tun)
2050                 return -EBADFD;
2051
2052         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2053                 noblock = 1;
2054
2055         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2056
2057         tun_put(tun);
2058         return result;
2059 }
2060
2061 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2062                                 struct tun_file *tfile,
2063                                 struct xdp_frame *xdp_frame,
2064                                 struct iov_iter *iter)
2065 {
2066         int vnet_hdr_sz = 0;
2067         size_t size = xdp_frame->len;
2068         size_t ret;
2069
2070         if (tun->flags & IFF_VNET_HDR) {
2071                 struct virtio_net_hdr gso = { 0 };
2072
2073                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2074                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2075                         return -EINVAL;
2076                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2077                              sizeof(gso)))
2078                         return -EFAULT;
2079                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2080         }
2081
2082         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2083
2084         preempt_disable();
2085         dev_sw_netstats_tx_add(tun->dev, 1, ret);
2086         preempt_enable();
2087
2088         return ret;
2089 }
2090
2091 /* Put packet to the user space buffer */
2092 static ssize_t tun_put_user(struct tun_struct *tun,
2093                             struct tun_file *tfile,
2094                             struct sk_buff *skb,
2095                             struct iov_iter *iter)
2096 {
2097         struct tun_pi pi = { 0, skb->protocol };
2098         ssize_t total;
2099         int vlan_offset = 0;
2100         int vlan_hlen = 0;
2101         int vnet_hdr_sz = 0;
2102
2103         if (skb_vlan_tag_present(skb))
2104                 vlan_hlen = VLAN_HLEN;
2105
2106         if (tun->flags & IFF_VNET_HDR)
2107                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2108
2109         total = skb->len + vlan_hlen + vnet_hdr_sz;
2110
2111         if (!(tun->flags & IFF_NO_PI)) {
2112                 if (iov_iter_count(iter) < sizeof(pi))
2113                         return -EINVAL;
2114
2115                 total += sizeof(pi);
2116                 if (iov_iter_count(iter) < total) {
2117                         /* Packet will be striped */
2118                         pi.flags |= TUN_PKT_STRIP;
2119                 }
2120
2121                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2122                         return -EFAULT;
2123         }
2124
2125         if (vnet_hdr_sz) {
2126                 struct virtio_net_hdr gso;
2127
2128                 if (iov_iter_count(iter) < vnet_hdr_sz)
2129                         return -EINVAL;
2130
2131                 if (virtio_net_hdr_from_skb(skb, &gso,
2132                                             tun_is_little_endian(tun), true,
2133                                             vlan_hlen)) {
2134                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2135
2136                         if (net_ratelimit()) {
2137                                 netdev_err(tun->dev, "unexpected GSO type: 0x%x, gso_size %d, hdr_len %d\n",
2138                                            sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2139                                            tun16_to_cpu(tun, gso.hdr_len));
2140                                 print_hex_dump(KERN_ERR, "tun: ",
2141                                                DUMP_PREFIX_NONE,
2142                                                16, 1, skb->head,
2143                                                min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2144                         }
2145                         WARN_ON_ONCE(1);
2146                         return -EINVAL;
2147                 }
2148
2149                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2150                         return -EFAULT;
2151
2152                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2153         }
2154
2155         if (vlan_hlen) {
2156                 int ret;
2157                 struct veth veth;
2158
2159                 veth.h_vlan_proto = skb->vlan_proto;
2160                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2161
2162                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2163
2164                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2165                 if (ret || !iov_iter_count(iter))
2166                         goto done;
2167
2168                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2169                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2170                         goto done;
2171         }
2172
2173         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2174
2175 done:
2176         /* caller is in process context, */
2177         preempt_disable();
2178         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2179         preempt_enable();
2180
2181         return total;
2182 }
2183
2184 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2185 {
2186         DECLARE_WAITQUEUE(wait, current);
2187         void *ptr = NULL;
2188         int error = 0;
2189
2190         ptr = ptr_ring_consume(&tfile->tx_ring);
2191         if (ptr)
2192                 goto out;
2193         if (noblock) {
2194                 error = -EAGAIN;
2195                 goto out;
2196         }
2197
2198         add_wait_queue(&tfile->socket.wq.wait, &wait);
2199
2200         while (1) {
2201                 set_current_state(TASK_INTERRUPTIBLE);
2202                 ptr = ptr_ring_consume(&tfile->tx_ring);
2203                 if (ptr)
2204                         break;
2205                 if (signal_pending(current)) {
2206                         error = -ERESTARTSYS;
2207                         break;
2208                 }
2209                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2210                         error = -EFAULT;
2211                         break;
2212                 }
2213
2214                 schedule();
2215         }
2216
2217         __set_current_state(TASK_RUNNING);
2218         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2219
2220 out:
2221         *err = error;
2222         return ptr;
2223 }
2224
2225 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2226                            struct iov_iter *to,
2227                            int noblock, void *ptr)
2228 {
2229         ssize_t ret;
2230         int err;
2231
2232         if (!iov_iter_count(to)) {
2233                 tun_ptr_free(ptr);
2234                 return 0;
2235         }
2236
2237         if (!ptr) {
2238                 /* Read frames from ring */
2239                 ptr = tun_ring_recv(tfile, noblock, &err);
2240                 if (!ptr)
2241                         return err;
2242         }
2243
2244         if (tun_is_xdp_frame(ptr)) {
2245                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2246
2247                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2248                 xdp_return_frame(xdpf);
2249         } else {
2250                 struct sk_buff *skb = ptr;
2251
2252                 ret = tun_put_user(tun, tfile, skb, to);
2253                 if (unlikely(ret < 0))
2254                         kfree_skb(skb);
2255                 else
2256                         consume_skb(skb);
2257         }
2258
2259         return ret;
2260 }
2261
2262 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2263 {
2264         struct file *file = iocb->ki_filp;
2265         struct tun_file *tfile = file->private_data;
2266         struct tun_struct *tun = tun_get(tfile);
2267         ssize_t len = iov_iter_count(to), ret;
2268         int noblock = 0;
2269
2270         if (!tun)
2271                 return -EBADFD;
2272
2273         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2274                 noblock = 1;
2275
2276         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2277         ret = min_t(ssize_t, ret, len);
2278         if (ret > 0)
2279                 iocb->ki_pos = ret;
2280         tun_put(tun);
2281         return ret;
2282 }
2283
2284 static void tun_prog_free(struct rcu_head *rcu)
2285 {
2286         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2287
2288         bpf_prog_destroy(prog->prog);
2289         kfree(prog);
2290 }
2291
2292 static int __tun_set_ebpf(struct tun_struct *tun,
2293                           struct tun_prog __rcu **prog_p,
2294                           struct bpf_prog *prog)
2295 {
2296         struct tun_prog *old, *new = NULL;
2297
2298         if (prog) {
2299                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2300                 if (!new)
2301                         return -ENOMEM;
2302                 new->prog = prog;
2303         }
2304
2305         spin_lock_bh(&tun->lock);
2306         old = rcu_dereference_protected(*prog_p,
2307                                         lockdep_is_held(&tun->lock));
2308         rcu_assign_pointer(*prog_p, new);
2309         spin_unlock_bh(&tun->lock);
2310
2311         if (old)
2312                 call_rcu(&old->rcu, tun_prog_free);
2313
2314         return 0;
2315 }
2316
2317 static void tun_free_netdev(struct net_device *dev)
2318 {
2319         struct tun_struct *tun = netdev_priv(dev);
2320
2321         BUG_ON(!(list_empty(&tun->disabled)));
2322
2323         free_percpu(dev->tstats);
2324         tun_flow_uninit(tun);
2325         security_tun_dev_free_security(tun->security);
2326         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2327         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2328 }
2329
2330 static void tun_setup(struct net_device *dev)
2331 {
2332         struct tun_struct *tun = netdev_priv(dev);
2333
2334         tun->owner = INVALID_UID;
2335         tun->group = INVALID_GID;
2336         tun_default_link_ksettings(dev, &tun->link_ksettings);
2337
2338         dev->ethtool_ops = &tun_ethtool_ops;
2339         dev->needs_free_netdev = true;
2340         dev->priv_destructor = tun_free_netdev;
2341         /* We prefer our own queue length */
2342         dev->tx_queue_len = TUN_READQ_SIZE;
2343 }
2344
2345 /* Trivial set of netlink ops to allow deleting tun or tap
2346  * device with netlink.
2347  */
2348 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2349                         struct netlink_ext_ack *extack)
2350 {
2351         NL_SET_ERR_MSG(extack,
2352                        "tun/tap creation via rtnetlink is not supported.");
2353         return -EOPNOTSUPP;
2354 }
2355
2356 static size_t tun_get_size(const struct net_device *dev)
2357 {
2358         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2359         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2360
2361         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2362                nla_total_size(sizeof(gid_t)) + /* GROUP */
2363                nla_total_size(sizeof(u8)) + /* TYPE */
2364                nla_total_size(sizeof(u8)) + /* PI */
2365                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2366                nla_total_size(sizeof(u8)) + /* PERSIST */
2367                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2368                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2369                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2370                0;
2371 }
2372
2373 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2374 {
2375         struct tun_struct *tun = netdev_priv(dev);
2376
2377         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2378                 goto nla_put_failure;
2379         if (uid_valid(tun->owner) &&
2380             nla_put_u32(skb, IFLA_TUN_OWNER,
2381                         from_kuid_munged(current_user_ns(), tun->owner)))
2382                 goto nla_put_failure;
2383         if (gid_valid(tun->group) &&
2384             nla_put_u32(skb, IFLA_TUN_GROUP,
2385                         from_kgid_munged(current_user_ns(), tun->group)))
2386                 goto nla_put_failure;
2387         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2388                 goto nla_put_failure;
2389         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2390                 goto nla_put_failure;
2391         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2392                 goto nla_put_failure;
2393         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2394                        !!(tun->flags & IFF_MULTI_QUEUE)))
2395                 goto nla_put_failure;
2396         if (tun->flags & IFF_MULTI_QUEUE) {
2397                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2398                         goto nla_put_failure;
2399                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2400                                 tun->numdisabled))
2401                         goto nla_put_failure;
2402         }
2403
2404         return 0;
2405
2406 nla_put_failure:
2407         return -EMSGSIZE;
2408 }
2409
2410 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2411         .kind           = DRV_NAME,
2412         .priv_size      = sizeof(struct tun_struct),
2413         .setup          = tun_setup,
2414         .validate       = tun_validate,
2415         .get_size       = tun_get_size,
2416         .fill_info      = tun_fill_info,
2417 };
2418
2419 static void tun_sock_write_space(struct sock *sk)
2420 {
2421         struct tun_file *tfile;
2422         wait_queue_head_t *wqueue;
2423
2424         if (!sock_writeable(sk))
2425                 return;
2426
2427         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2428                 return;
2429
2430         wqueue = sk_sleep(sk);
2431         if (wqueue && waitqueue_active(wqueue))
2432                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2433                                                 EPOLLWRNORM | EPOLLWRBAND);
2434
2435         tfile = container_of(sk, struct tun_file, sk);
2436         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2437 }
2438
2439 static void tun_put_page(struct tun_page *tpage)
2440 {
2441         if (tpage->page)
2442                 __page_frag_cache_drain(tpage->page, tpage->count);
2443 }
2444
2445 static int tun_xdp_one(struct tun_struct *tun,
2446                        struct tun_file *tfile,
2447                        struct xdp_buff *xdp, int *flush,
2448                        struct tun_page *tpage)
2449 {
2450         unsigned int datasize = xdp->data_end - xdp->data;
2451         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2452         struct virtio_net_hdr *gso = &hdr->gso;
2453         struct bpf_prog *xdp_prog;
2454         struct sk_buff *skb = NULL;
2455         struct sk_buff_head *queue;
2456         u32 rxhash = 0, act;
2457         int buflen = hdr->buflen;
2458         int ret = 0;
2459         bool skb_xdp = false;
2460         struct page *page;
2461
2462         xdp_prog = rcu_dereference(tun->xdp_prog);
2463         if (xdp_prog) {
2464                 if (gso->gso_type) {
2465                         skb_xdp = true;
2466                         goto build;
2467                 }
2468
2469                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2470                 xdp_set_data_meta_invalid(xdp);
2471
2472                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2473                 ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2474                 if (ret < 0) {
2475                         put_page(virt_to_head_page(xdp->data));
2476                         return ret;
2477                 }
2478
2479                 switch (ret) {
2480                 case XDP_REDIRECT:
2481                         *flush = true;
2482                         fallthrough;
2483                 case XDP_TX:
2484                         return 0;
2485                 case XDP_PASS:
2486                         break;
2487                 default:
2488                         page = virt_to_head_page(xdp->data);
2489                         if (tpage->page == page) {
2490                                 ++tpage->count;
2491                         } else {
2492                                 tun_put_page(tpage);
2493                                 tpage->page = page;
2494                                 tpage->count = 1;
2495                         }
2496                         return 0;
2497                 }
2498         }
2499
2500 build:
2501         skb = build_skb(xdp->data_hard_start, buflen);
2502         if (!skb) {
2503                 ret = -ENOMEM;
2504                 goto out;
2505         }
2506
2507         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2508         skb_put(skb, xdp->data_end - xdp->data);
2509
2510         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2511                 atomic_long_inc(&tun->rx_frame_errors);
2512                 kfree_skb(skb);
2513                 ret = -EINVAL;
2514                 goto out;
2515         }
2516
2517         skb->protocol = eth_type_trans(skb, tun->dev);
2518         skb_reset_network_header(skb);
2519         skb_probe_transport_header(skb);
2520         skb_record_rx_queue(skb, tfile->queue_index);
2521
2522         if (skb_xdp) {
2523                 ret = do_xdp_generic(xdp_prog, skb);
2524                 if (ret != XDP_PASS) {
2525                         ret = 0;
2526                         goto out;
2527                 }
2528         }
2529
2530         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2531             !tfile->detached)
2532                 rxhash = __skb_get_hash_symmetric(skb);
2533
2534         if (tfile->napi_enabled) {
2535                 queue = &tfile->sk.sk_write_queue;
2536                 spin_lock(&queue->lock);
2537
2538                 if (unlikely(tfile->detached)) {
2539                         spin_unlock(&queue->lock);
2540                         kfree_skb(skb);
2541                         return -EBUSY;
2542                 }
2543
2544                 __skb_queue_tail(queue, skb);
2545                 spin_unlock(&queue->lock);
2546                 ret = 1;
2547         } else {
2548                 netif_receive_skb(skb);
2549                 ret = 0;
2550         }
2551
2552         /* No need to disable preemption here since this function is
2553          * always called with bh disabled
2554          */
2555         dev_sw_netstats_rx_add(tun->dev, datasize);
2556
2557         if (rxhash)
2558                 tun_flow_update(tun, rxhash, tfile);
2559
2560 out:
2561         return ret;
2562 }
2563
2564 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2565 {
2566         int ret, i;
2567         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2568         struct tun_struct *tun = tun_get(tfile);
2569         struct tun_msg_ctl *ctl = m->msg_control;
2570         struct xdp_buff *xdp;
2571
2572         if (!tun)
2573                 return -EBADFD;
2574
2575         if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2576             ctl && ctl->type == TUN_MSG_PTR) {
2577                 struct tun_page tpage;
2578                 int n = ctl->num;
2579                 int flush = 0, queued = 0;
2580
2581                 memset(&tpage, 0, sizeof(tpage));
2582
2583                 local_bh_disable();
2584                 rcu_read_lock();
2585
2586                 for (i = 0; i < n; i++) {
2587                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2588                         ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2589                         if (ret > 0)
2590                                 queued += ret;
2591                 }
2592
2593                 if (flush)
2594                         xdp_do_flush();
2595
2596                 if (tfile->napi_enabled && queued > 0)
2597                         napi_schedule(&tfile->napi);
2598
2599                 rcu_read_unlock();
2600                 local_bh_enable();
2601
2602                 tun_put_page(&tpage);
2603
2604                 ret = total_len;
2605                 goto out;
2606         }
2607
2608         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2609                            m->msg_flags & MSG_DONTWAIT,
2610                            m->msg_flags & MSG_MORE);
2611 out:
2612         tun_put(tun);
2613         return ret;
2614 }
2615
2616 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2617                        int flags)
2618 {
2619         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2620         struct tun_struct *tun = tun_get(tfile);
2621         void *ptr = m->msg_control;
2622         int ret;
2623
2624         if (!tun) {
2625                 ret = -EBADFD;
2626                 goto out_free;
2627         }
2628
2629         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2630                 ret = -EINVAL;
2631                 goto out_put_tun;
2632         }
2633         if (flags & MSG_ERRQUEUE) {
2634                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2635                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2636                 goto out;
2637         }
2638         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2639         if (ret > (ssize_t)total_len) {
2640                 m->msg_flags |= MSG_TRUNC;
2641                 ret = flags & MSG_TRUNC ? ret : total_len;
2642         }
2643 out:
2644         tun_put(tun);
2645         return ret;
2646
2647 out_put_tun:
2648         tun_put(tun);
2649 out_free:
2650         tun_ptr_free(ptr);
2651         return ret;
2652 }
2653
2654 static int tun_ptr_peek_len(void *ptr)
2655 {
2656         if (likely(ptr)) {
2657                 if (tun_is_xdp_frame(ptr)) {
2658                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2659
2660                         return xdpf->len;
2661                 }
2662                 return __skb_array_len_with_tag(ptr);
2663         } else {
2664                 return 0;
2665         }
2666 }
2667
2668 static int tun_peek_len(struct socket *sock)
2669 {
2670         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2671         struct tun_struct *tun;
2672         int ret = 0;
2673
2674         tun = tun_get(tfile);
2675         if (!tun)
2676                 return 0;
2677
2678         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2679         tun_put(tun);
2680
2681         return ret;
2682 }
2683
2684 /* Ops structure to mimic raw sockets with tun */
2685 static const struct proto_ops tun_socket_ops = {
2686         .peek_len = tun_peek_len,
2687         .sendmsg = tun_sendmsg,
2688         .recvmsg = tun_recvmsg,
2689 };
2690
2691 static struct proto tun_proto = {
2692         .name           = "tun",
2693         .owner          = THIS_MODULE,
2694         .obj_size       = sizeof(struct tun_file),
2695 };
2696
2697 static int tun_flags(struct tun_struct *tun)
2698 {
2699         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2700 }
2701
2702 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2703                               char *buf)
2704 {
2705         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2706         return sysfs_emit(buf, "0x%x\n", tun_flags(tun));
2707 }
2708
2709 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2710                           char *buf)
2711 {
2712         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2713         return uid_valid(tun->owner)?
2714                 sysfs_emit(buf, "%u\n",
2715                            from_kuid_munged(current_user_ns(), tun->owner)) :
2716                 sysfs_emit(buf, "-1\n");
2717 }
2718
2719 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2720                           char *buf)
2721 {
2722         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2723         return gid_valid(tun->group) ?
2724                 sysfs_emit(buf, "%u\n",
2725                            from_kgid_munged(current_user_ns(), tun->group)) :
2726                 sysfs_emit(buf, "-1\n");
2727 }
2728
2729 static DEVICE_ATTR_RO(tun_flags);
2730 static DEVICE_ATTR_RO(owner);
2731 static DEVICE_ATTR_RO(group);
2732
2733 static struct attribute *tun_dev_attrs[] = {
2734         &dev_attr_tun_flags.attr,
2735         &dev_attr_owner.attr,
2736         &dev_attr_group.attr,
2737         NULL
2738 };
2739
2740 static const struct attribute_group tun_attr_group = {
2741         .attrs = tun_dev_attrs
2742 };
2743
2744 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2745 {
2746         struct tun_struct *tun;
2747         struct tun_file *tfile = file->private_data;
2748         struct net_device *dev;
2749         int err;
2750
2751         if (tfile->detached)
2752                 return -EINVAL;
2753
2754         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2755                 if (!capable(CAP_NET_ADMIN))
2756                         return -EPERM;
2757
2758                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2759                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2760                         return -EINVAL;
2761         }
2762
2763         dev = __dev_get_by_name(net, ifr->ifr_name);
2764         if (dev) {
2765                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2766                         return -EBUSY;
2767                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2768                         tun = netdev_priv(dev);
2769                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2770                         tun = netdev_priv(dev);
2771                 else
2772                         return -EINVAL;
2773
2774                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2775                     !!(tun->flags & IFF_MULTI_QUEUE))
2776                         return -EINVAL;
2777
2778                 if (tun_not_capable(tun))
2779                         return -EPERM;
2780                 err = security_tun_dev_open(tun->security);
2781                 if (err < 0)
2782                         return err;
2783
2784                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2785                                  ifr->ifr_flags & IFF_NAPI,
2786                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2787                 if (err < 0)
2788                         return err;
2789
2790                 if (tun->flags & IFF_MULTI_QUEUE &&
2791                     (tun->numqueues + tun->numdisabled > 1)) {
2792                         /* One or more queue has already been attached, no need
2793                          * to initialize the device again.
2794                          */
2795                         netdev_state_change(dev);
2796                         return 0;
2797                 }
2798
2799                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2800                               (ifr->ifr_flags & TUN_FEATURES);
2801
2802                 netdev_state_change(dev);
2803         } else {
2804                 char *name;
2805                 unsigned long flags = 0;
2806                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2807                              MAX_TAP_QUEUES : 1;
2808
2809                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2810                         return -EPERM;
2811                 err = security_tun_dev_create();
2812                 if (err < 0)
2813                         return err;
2814
2815                 /* Set dev type */
2816                 if (ifr->ifr_flags & IFF_TUN) {
2817                         /* TUN device */
2818                         flags |= IFF_TUN;
2819                         name = "tun%d";
2820                 } else if (ifr->ifr_flags & IFF_TAP) {
2821                         /* TAP device */
2822                         flags |= IFF_TAP;
2823                         name = "tap%d";
2824                 } else
2825                         return -EINVAL;
2826
2827                 if (*ifr->ifr_name)
2828                         name = ifr->ifr_name;
2829
2830                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2831                                        NET_NAME_UNKNOWN, tun_setup, queues,
2832                                        queues);
2833
2834                 if (!dev)
2835                         return -ENOMEM;
2836
2837                 dev_net_set(dev, net);
2838                 dev->rtnl_link_ops = &tun_link_ops;
2839                 dev->ifindex = tfile->ifindex;
2840                 dev->sysfs_groups[0] = &tun_attr_group;
2841
2842                 tun = netdev_priv(dev);
2843                 tun->dev = dev;
2844                 tun->flags = flags;
2845                 tun->txflt.count = 0;
2846                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2847
2848                 tun->align = NET_SKB_PAD;
2849                 tun->filter_attached = false;
2850                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2851                 tun->rx_batched = 0;
2852                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2853
2854                 tun->ifr = ifr;
2855                 tun->file = file;
2856
2857                 tun_net_initialize(dev);
2858
2859                 err = register_netdevice(tun->dev);
2860                 if (err < 0) {
2861                         free_netdev(dev);
2862                         return err;
2863                 }
2864                 /* free_netdev() won't check refcnt, to avoid race
2865                  * with dev_put() we need publish tun after registration.
2866                  */
2867                 rcu_assign_pointer(tfile->tun, tun);
2868         }
2869
2870         if (ifr->ifr_flags & IFF_NO_CARRIER)
2871                 netif_carrier_off(tun->dev);
2872         else
2873                 netif_carrier_on(tun->dev);
2874
2875         /* Make sure persistent devices do not get stuck in
2876          * xoff state.
2877          */
2878         if (netif_running(tun->dev))
2879                 netif_tx_wake_all_queues(tun->dev);
2880
2881         strcpy(ifr->ifr_name, tun->dev->name);
2882         return 0;
2883 }
2884
2885 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2886 {
2887         strcpy(ifr->ifr_name, tun->dev->name);
2888
2889         ifr->ifr_flags = tun_flags(tun);
2890
2891 }
2892
2893 /* This is like a cut-down ethtool ops, except done via tun fd so no
2894  * privs required. */
2895 static int set_offload(struct tun_struct *tun, unsigned long arg)
2896 {
2897         netdev_features_t features = 0;
2898
2899         if (arg & TUN_F_CSUM) {
2900                 features |= NETIF_F_HW_CSUM;
2901                 arg &= ~TUN_F_CSUM;
2902
2903                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2904                         if (arg & TUN_F_TSO_ECN) {
2905                                 features |= NETIF_F_TSO_ECN;
2906                                 arg &= ~TUN_F_TSO_ECN;
2907                         }
2908                         if (arg & TUN_F_TSO4)
2909                                 features |= NETIF_F_TSO;
2910                         if (arg & TUN_F_TSO6)
2911                                 features |= NETIF_F_TSO6;
2912                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2913                 }
2914
2915                 arg &= ~TUN_F_UFO;
2916
2917                 /* TODO: for now USO4 and USO6 should work simultaneously */
2918                 if (arg & TUN_F_USO4 && arg & TUN_F_USO6) {
2919                         features |= NETIF_F_GSO_UDP_L4;
2920                         arg &= ~(TUN_F_USO4 | TUN_F_USO6);
2921                 }
2922         }
2923
2924         /* This gives the user a way to test for new features in future by
2925          * trying to set them. */
2926         if (arg)
2927                 return -EINVAL;
2928
2929         tun->set_features = features;
2930         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2931         tun->dev->wanted_features |= features;
2932         netdev_update_features(tun->dev);
2933
2934         return 0;
2935 }
2936
2937 static void tun_detach_filter(struct tun_struct *tun, int n)
2938 {
2939         int i;
2940         struct tun_file *tfile;
2941
2942         for (i = 0; i < n; i++) {
2943                 tfile = rtnl_dereference(tun->tfiles[i]);
2944                 lock_sock(tfile->socket.sk);
2945                 sk_detach_filter(tfile->socket.sk);
2946                 release_sock(tfile->socket.sk);
2947         }
2948
2949         tun->filter_attached = false;
2950 }
2951
2952 static int tun_attach_filter(struct tun_struct *tun)
2953 {
2954         int i, ret = 0;
2955         struct tun_file *tfile;
2956
2957         for (i = 0; i < tun->numqueues; i++) {
2958                 tfile = rtnl_dereference(tun->tfiles[i]);
2959                 lock_sock(tfile->socket.sk);
2960                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2961                 release_sock(tfile->socket.sk);
2962                 if (ret) {
2963                         tun_detach_filter(tun, i);
2964                         return ret;
2965                 }
2966         }
2967
2968         tun->filter_attached = true;
2969         return ret;
2970 }
2971
2972 static void tun_set_sndbuf(struct tun_struct *tun)
2973 {
2974         struct tun_file *tfile;
2975         int i;
2976
2977         for (i = 0; i < tun->numqueues; i++) {
2978                 tfile = rtnl_dereference(tun->tfiles[i]);
2979                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2980         }
2981 }
2982
2983 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2984 {
2985         struct tun_file *tfile = file->private_data;
2986         struct tun_struct *tun;
2987         int ret = 0;
2988
2989         rtnl_lock();
2990
2991         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2992                 tun = tfile->detached;
2993                 if (!tun) {
2994                         ret = -EINVAL;
2995                         goto unlock;
2996                 }
2997                 ret = security_tun_dev_attach_queue(tun->security);
2998                 if (ret < 0)
2999                         goto unlock;
3000                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
3001                                  tun->flags & IFF_NAPI_FRAGS, true);
3002         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
3003                 tun = rtnl_dereference(tfile->tun);
3004                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
3005                         ret = -EINVAL;
3006                 else
3007                         __tun_detach(tfile, false);
3008         } else
3009                 ret = -EINVAL;
3010
3011         if (ret >= 0)
3012                 netdev_state_change(tun->dev);
3013
3014 unlock:
3015         rtnl_unlock();
3016         return ret;
3017 }
3018
3019 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
3020                         void __user *data)
3021 {
3022         struct bpf_prog *prog;
3023         int fd;
3024
3025         if (copy_from_user(&fd, data, sizeof(fd)))
3026                 return -EFAULT;
3027
3028         if (fd == -1) {
3029                 prog = NULL;
3030         } else {
3031                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3032                 if (IS_ERR(prog))
3033                         return PTR_ERR(prog);
3034         }
3035
3036         return __tun_set_ebpf(tun, prog_p, prog);
3037 }
3038
3039 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3040 static unsigned char tun_get_addr_len(unsigned short type)
3041 {
3042         switch (type) {
3043         case ARPHRD_IP6GRE:
3044         case ARPHRD_TUNNEL6:
3045                 return sizeof(struct in6_addr);
3046         case ARPHRD_IPGRE:
3047         case ARPHRD_TUNNEL:
3048         case ARPHRD_SIT:
3049                 return 4;
3050         case ARPHRD_ETHER:
3051                 return ETH_ALEN;
3052         case ARPHRD_IEEE802154:
3053         case ARPHRD_IEEE802154_MONITOR:
3054                 return IEEE802154_EXTENDED_ADDR_LEN;
3055         case ARPHRD_PHONET_PIPE:
3056         case ARPHRD_PPP:
3057         case ARPHRD_NONE:
3058                 return 0;
3059         case ARPHRD_6LOWPAN:
3060                 return EUI64_ADDR_LEN;
3061         case ARPHRD_FDDI:
3062                 return FDDI_K_ALEN;
3063         case ARPHRD_HIPPI:
3064                 return HIPPI_ALEN;
3065         case ARPHRD_IEEE802:
3066                 return FC_ALEN;
3067         case ARPHRD_ROSE:
3068                 return ROSE_ADDR_LEN;
3069         case ARPHRD_NETROM:
3070                 return AX25_ADDR_LEN;
3071         case ARPHRD_LOCALTLK:
3072                 return LTALK_ALEN;
3073         default:
3074                 return 0;
3075         }
3076 }
3077
3078 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3079                             unsigned long arg, int ifreq_len)
3080 {
3081         struct tun_file *tfile = file->private_data;
3082         struct net *net = sock_net(&tfile->sk);
3083         struct tun_struct *tun;
3084         void __user* argp = (void __user*)arg;
3085         unsigned int carrier;
3086         struct ifreq ifr;
3087         kuid_t owner;
3088         kgid_t group;
3089         int ifindex;
3090         int sndbuf;
3091         int vnet_hdr_sz;
3092         int le;
3093         int ret;
3094         bool do_notify = false;
3095
3096         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3097             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3098                 if (copy_from_user(&ifr, argp, ifreq_len))
3099                         return -EFAULT;
3100         } else {
3101                 memset(&ifr, 0, sizeof(ifr));
3102         }
3103         if (cmd == TUNGETFEATURES) {
3104                 /* Currently this just means: "what IFF flags are valid?".
3105                  * This is needed because we never checked for invalid flags on
3106                  * TUNSETIFF.
3107                  */
3108                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_CARRIER |
3109                                 TUN_FEATURES, (unsigned int __user*)argp);
3110         } else if (cmd == TUNSETQUEUE) {
3111                 return tun_set_queue(file, &ifr);
3112         } else if (cmd == SIOCGSKNS) {
3113                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3114                         return -EPERM;
3115                 return open_related_ns(&net->ns, get_net_ns);
3116         }
3117
3118         rtnl_lock();
3119
3120         tun = tun_get(tfile);
3121         if (cmd == TUNSETIFF) {
3122                 ret = -EEXIST;
3123                 if (tun)
3124                         goto unlock;
3125
3126                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3127
3128                 ret = tun_set_iff(net, file, &ifr);
3129
3130                 if (ret)
3131                         goto unlock;
3132
3133                 if (copy_to_user(argp, &ifr, ifreq_len))
3134                         ret = -EFAULT;
3135                 goto unlock;
3136         }
3137         if (cmd == TUNSETIFINDEX) {
3138                 ret = -EPERM;
3139                 if (tun)
3140                         goto unlock;
3141
3142                 ret = -EFAULT;
3143                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3144                         goto unlock;
3145                 ret = -EINVAL;
3146                 if (ifindex < 0)
3147                         goto unlock;
3148                 ret = 0;
3149                 tfile->ifindex = ifindex;
3150                 goto unlock;
3151         }
3152
3153         ret = -EBADFD;
3154         if (!tun)
3155                 goto unlock;
3156
3157         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3158
3159         net = dev_net(tun->dev);
3160         ret = 0;
3161         switch (cmd) {
3162         case TUNGETIFF:
3163                 tun_get_iff(tun, &ifr);
3164
3165                 if (tfile->detached)
3166                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3167                 if (!tfile->socket.sk->sk_filter)
3168                         ifr.ifr_flags |= IFF_NOFILTER;
3169
3170                 if (copy_to_user(argp, &ifr, ifreq_len))
3171                         ret = -EFAULT;
3172                 break;
3173
3174         case TUNSETNOCSUM:
3175                 /* Disable/Enable checksum */
3176
3177                 /* [unimplemented] */
3178                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3179                            arg ? "disabled" : "enabled");
3180                 break;
3181
3182         case TUNSETPERSIST:
3183                 /* Disable/Enable persist mode. Keep an extra reference to the
3184                  * module to prevent the module being unprobed.
3185                  */
3186                 if (arg && !(tun->flags & IFF_PERSIST)) {
3187                         tun->flags |= IFF_PERSIST;
3188                         __module_get(THIS_MODULE);
3189                         do_notify = true;
3190                 }
3191                 if (!arg && (tun->flags & IFF_PERSIST)) {
3192                         tun->flags &= ~IFF_PERSIST;
3193                         module_put(THIS_MODULE);
3194                         do_notify = true;
3195                 }
3196
3197                 netif_info(tun, drv, tun->dev, "persist %s\n",
3198                            arg ? "enabled" : "disabled");
3199                 break;
3200
3201         case TUNSETOWNER:
3202                 /* Set owner of the device */
3203                 owner = make_kuid(current_user_ns(), arg);
3204                 if (!uid_valid(owner)) {
3205                         ret = -EINVAL;
3206                         break;
3207                 }
3208                 tun->owner = owner;
3209                 do_notify = true;
3210                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3211                            from_kuid(&init_user_ns, tun->owner));
3212                 break;
3213
3214         case TUNSETGROUP:
3215                 /* Set group of the device */
3216                 group = make_kgid(current_user_ns(), arg);
3217                 if (!gid_valid(group)) {
3218                         ret = -EINVAL;
3219                         break;
3220                 }
3221                 tun->group = group;
3222                 do_notify = true;
3223                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3224                            from_kgid(&init_user_ns, tun->group));
3225                 break;
3226
3227         case TUNSETLINK:
3228                 /* Only allow setting the type when the interface is down */
3229                 if (tun->dev->flags & IFF_UP) {
3230                         netif_info(tun, drv, tun->dev,
3231                                    "Linktype set failed because interface is up\n");
3232                         ret = -EBUSY;
3233                 } else {
3234                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3235                                                        tun->dev);
3236                         ret = notifier_to_errno(ret);
3237                         if (ret) {
3238                                 netif_info(tun, drv, tun->dev,
3239                                            "Refused to change device type\n");
3240                                 break;
3241                         }
3242                         tun->dev->type = (int) arg;
3243                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3244                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3245                                    tun->dev->type);
3246                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3247                                                  tun->dev);
3248                 }
3249                 break;
3250
3251         case TUNSETDEBUG:
3252                 tun->msg_enable = (u32)arg;
3253                 break;
3254
3255         case TUNSETOFFLOAD:
3256                 ret = set_offload(tun, arg);
3257                 break;
3258
3259         case TUNSETTXFILTER:
3260                 /* Can be set only for TAPs */
3261                 ret = -EINVAL;
3262                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3263                         break;
3264                 ret = update_filter(&tun->txflt, (void __user *)arg);
3265                 break;
3266
3267         case SIOCGIFHWADDR:
3268                 /* Get hw address */
3269                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3270                 if (copy_to_user(argp, &ifr, ifreq_len))
3271                         ret = -EFAULT;
3272                 break;
3273
3274         case SIOCSIFHWADDR:
3275                 /* Set hw address */
3276                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3277                 break;
3278
3279         case TUNGETSNDBUF:
3280                 sndbuf = tfile->socket.sk->sk_sndbuf;
3281                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3282                         ret = -EFAULT;
3283                 break;
3284
3285         case TUNSETSNDBUF:
3286                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3287                         ret = -EFAULT;
3288                         break;
3289                 }
3290                 if (sndbuf <= 0) {
3291                         ret = -EINVAL;
3292                         break;
3293                 }
3294
3295                 tun->sndbuf = sndbuf;
3296                 tun_set_sndbuf(tun);
3297                 break;
3298
3299         case TUNGETVNETHDRSZ:
3300                 vnet_hdr_sz = tun->vnet_hdr_sz;
3301                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3302                         ret = -EFAULT;
3303                 break;
3304
3305         case TUNSETVNETHDRSZ:
3306                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3307                         ret = -EFAULT;
3308                         break;
3309                 }
3310                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3311                         ret = -EINVAL;
3312                         break;
3313                 }
3314
3315                 tun->vnet_hdr_sz = vnet_hdr_sz;
3316                 break;
3317
3318         case TUNGETVNETLE:
3319                 le = !!(tun->flags & TUN_VNET_LE);
3320                 if (put_user(le, (int __user *)argp))
3321                         ret = -EFAULT;
3322                 break;
3323
3324         case TUNSETVNETLE:
3325                 if (get_user(le, (int __user *)argp)) {
3326                         ret = -EFAULT;
3327                         break;
3328                 }
3329                 if (le)
3330                         tun->flags |= TUN_VNET_LE;
3331                 else
3332                         tun->flags &= ~TUN_VNET_LE;
3333                 break;
3334
3335         case TUNGETVNETBE:
3336                 ret = tun_get_vnet_be(tun, argp);
3337                 break;
3338
3339         case TUNSETVNETBE:
3340                 ret = tun_set_vnet_be(tun, argp);
3341                 break;
3342
3343         case TUNATTACHFILTER:
3344                 /* Can be set only for TAPs */
3345                 ret = -EINVAL;
3346                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3347                         break;
3348                 ret = -EFAULT;
3349                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3350                         break;
3351
3352                 ret = tun_attach_filter(tun);
3353                 break;
3354
3355         case TUNDETACHFILTER:
3356                 /* Can be set only for TAPs */
3357                 ret = -EINVAL;
3358                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3359                         break;
3360                 ret = 0;
3361                 tun_detach_filter(tun, tun->numqueues);
3362                 break;
3363
3364         case TUNGETFILTER:
3365                 ret = -EINVAL;
3366                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3367                         break;
3368                 ret = -EFAULT;
3369                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3370                         break;
3371                 ret = 0;
3372                 break;
3373
3374         case TUNSETSTEERINGEBPF:
3375                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3376                 break;
3377
3378         case TUNSETFILTEREBPF:
3379                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3380                 break;
3381
3382         case TUNSETCARRIER:
3383                 ret = -EFAULT;
3384                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3385                         goto unlock;
3386
3387                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3388                 break;
3389
3390         case TUNGETDEVNETNS:
3391                 ret = -EPERM;
3392                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3393                         goto unlock;
3394                 ret = open_related_ns(&net->ns, get_net_ns);
3395                 break;
3396
3397         default:
3398                 ret = -EINVAL;
3399                 break;
3400         }
3401
3402         if (do_notify)
3403                 netdev_state_change(tun->dev);
3404
3405 unlock:
3406         rtnl_unlock();
3407         if (tun)
3408                 tun_put(tun);
3409         return ret;
3410 }
3411
3412 static long tun_chr_ioctl(struct file *file,
3413                           unsigned int cmd, unsigned long arg)
3414 {
3415         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3416 }
3417
3418 #ifdef CONFIG_COMPAT
3419 static long tun_chr_compat_ioctl(struct file *file,
3420                          unsigned int cmd, unsigned long arg)
3421 {
3422         switch (cmd) {
3423         case TUNSETIFF:
3424         case TUNGETIFF:
3425         case TUNSETTXFILTER:
3426         case TUNGETSNDBUF:
3427         case TUNSETSNDBUF:
3428         case SIOCGIFHWADDR:
3429         case SIOCSIFHWADDR:
3430                 arg = (unsigned long)compat_ptr(arg);
3431                 break;
3432         default:
3433                 arg = (compat_ulong_t)arg;
3434                 break;
3435         }
3436
3437         /*
3438          * compat_ifreq is shorter than ifreq, so we must not access beyond
3439          * the end of that structure. All fields that are used in this
3440          * driver are compatible though, we don't need to convert the
3441          * contents.
3442          */
3443         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3444 }
3445 #endif /* CONFIG_COMPAT */
3446
3447 static int tun_chr_fasync(int fd, struct file *file, int on)
3448 {
3449         struct tun_file *tfile = file->private_data;
3450         int ret;
3451
3452         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3453                 goto out;
3454
3455         if (on) {
3456                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3457                 tfile->flags |= TUN_FASYNC;
3458         } else
3459                 tfile->flags &= ~TUN_FASYNC;
3460         ret = 0;
3461 out:
3462         return ret;
3463 }
3464
3465 static int tun_chr_open(struct inode *inode, struct file * file)
3466 {
3467         struct net *net = current->nsproxy->net_ns;
3468         struct tun_file *tfile;
3469
3470         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3471                                             &tun_proto, 0);
3472         if (!tfile)
3473                 return -ENOMEM;
3474         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3475                 sk_free(&tfile->sk);
3476                 return -ENOMEM;
3477         }
3478
3479         mutex_init(&tfile->napi_mutex);
3480         RCU_INIT_POINTER(tfile->tun, NULL);
3481         tfile->flags = 0;
3482         tfile->ifindex = 0;
3483
3484         init_waitqueue_head(&tfile->socket.wq.wait);
3485
3486         tfile->socket.file = file;
3487         tfile->socket.ops = &tun_socket_ops;
3488
3489         sock_init_data_uid(&tfile->socket, &tfile->sk, current_fsuid());
3490
3491         tfile->sk.sk_write_space = tun_sock_write_space;
3492         tfile->sk.sk_sndbuf = INT_MAX;
3493
3494         file->private_data = tfile;
3495         INIT_LIST_HEAD(&tfile->next);
3496
3497         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3498
3499         /* tun groks IOCB_NOWAIT just fine, mark it as such */
3500         file->f_mode |= FMODE_NOWAIT;
3501         return 0;
3502 }
3503
3504 static int tun_chr_close(struct inode *inode, struct file *file)
3505 {
3506         struct tun_file *tfile = file->private_data;
3507
3508         tun_detach(tfile, true);
3509
3510         return 0;
3511 }
3512
3513 #ifdef CONFIG_PROC_FS
3514 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3515 {
3516         struct tun_file *tfile = file->private_data;
3517         struct tun_struct *tun;
3518         struct ifreq ifr;
3519
3520         memset(&ifr, 0, sizeof(ifr));
3521
3522         rtnl_lock();
3523         tun = tun_get(tfile);
3524         if (tun)
3525                 tun_get_iff(tun, &ifr);
3526         rtnl_unlock();
3527
3528         if (tun)
3529                 tun_put(tun);
3530
3531         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3532 }
3533 #endif
3534
3535 static const struct file_operations tun_fops = {
3536         .owner  = THIS_MODULE,
3537         .llseek = no_llseek,
3538         .read_iter  = tun_chr_read_iter,
3539         .write_iter = tun_chr_write_iter,
3540         .poll   = tun_chr_poll,
3541         .unlocked_ioctl = tun_chr_ioctl,
3542 #ifdef CONFIG_COMPAT
3543         .compat_ioctl = tun_chr_compat_ioctl,
3544 #endif
3545         .open   = tun_chr_open,
3546         .release = tun_chr_close,
3547         .fasync = tun_chr_fasync,
3548 #ifdef CONFIG_PROC_FS
3549         .show_fdinfo = tun_chr_show_fdinfo,
3550 #endif
3551 };
3552
3553 static struct miscdevice tun_miscdev = {
3554         .minor = TUN_MINOR,
3555         .name = "tun",
3556         .nodename = "net/tun",
3557         .fops = &tun_fops,
3558 };
3559
3560 /* ethtool interface */
3561
3562 static void tun_default_link_ksettings(struct net_device *dev,
3563                                        struct ethtool_link_ksettings *cmd)
3564 {
3565         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3566         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3567         cmd->base.speed         = SPEED_10000;
3568         cmd->base.duplex        = DUPLEX_FULL;
3569         cmd->base.port          = PORT_TP;
3570         cmd->base.phy_address   = 0;
3571         cmd->base.autoneg       = AUTONEG_DISABLE;
3572 }
3573
3574 static int tun_get_link_ksettings(struct net_device *dev,
3575                                   struct ethtool_link_ksettings *cmd)
3576 {
3577         struct tun_struct *tun = netdev_priv(dev);
3578
3579         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3580         return 0;
3581 }
3582
3583 static int tun_set_link_ksettings(struct net_device *dev,
3584                                   const struct ethtool_link_ksettings *cmd)
3585 {
3586         struct tun_struct *tun = netdev_priv(dev);
3587
3588         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3589         return 0;
3590 }
3591
3592 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3593 {
3594         struct tun_struct *tun = netdev_priv(dev);
3595
3596         strscpy(info->driver, DRV_NAME, sizeof(info->driver));
3597         strscpy(info->version, DRV_VERSION, sizeof(info->version));
3598
3599         switch (tun->flags & TUN_TYPE_MASK) {
3600         case IFF_TUN:
3601                 strscpy(info->bus_info, "tun", sizeof(info->bus_info));
3602                 break;
3603         case IFF_TAP:
3604                 strscpy(info->bus_info, "tap", sizeof(info->bus_info));
3605                 break;
3606         }
3607 }
3608
3609 static u32 tun_get_msglevel(struct net_device *dev)
3610 {
3611         struct tun_struct *tun = netdev_priv(dev);
3612
3613         return tun->msg_enable;
3614 }
3615
3616 static void tun_set_msglevel(struct net_device *dev, u32 value)
3617 {
3618         struct tun_struct *tun = netdev_priv(dev);
3619
3620         tun->msg_enable = value;
3621 }
3622
3623 static int tun_get_coalesce(struct net_device *dev,
3624                             struct ethtool_coalesce *ec,
3625                             struct kernel_ethtool_coalesce *kernel_coal,
3626                             struct netlink_ext_ack *extack)
3627 {
3628         struct tun_struct *tun = netdev_priv(dev);
3629
3630         ec->rx_max_coalesced_frames = tun->rx_batched;
3631
3632         return 0;
3633 }
3634
3635 static int tun_set_coalesce(struct net_device *dev,
3636                             struct ethtool_coalesce *ec,
3637                             struct kernel_ethtool_coalesce *kernel_coal,
3638                             struct netlink_ext_ack *extack)
3639 {
3640         struct tun_struct *tun = netdev_priv(dev);
3641
3642         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3643                 tun->rx_batched = NAPI_POLL_WEIGHT;
3644         else
3645                 tun->rx_batched = ec->rx_max_coalesced_frames;
3646
3647         return 0;
3648 }
3649
3650 static const struct ethtool_ops tun_ethtool_ops = {
3651         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3652         .get_drvinfo    = tun_get_drvinfo,
3653         .get_msglevel   = tun_get_msglevel,
3654         .set_msglevel   = tun_set_msglevel,
3655         .get_link       = ethtool_op_get_link,
3656         .get_ts_info    = ethtool_op_get_ts_info,
3657         .get_coalesce   = tun_get_coalesce,
3658         .set_coalesce   = tun_set_coalesce,
3659         .get_link_ksettings = tun_get_link_ksettings,
3660         .set_link_ksettings = tun_set_link_ksettings,
3661 };
3662
3663 static int tun_queue_resize(struct tun_struct *tun)
3664 {
3665         struct net_device *dev = tun->dev;
3666         struct tun_file *tfile;
3667         struct ptr_ring **rings;
3668         int n = tun->numqueues + tun->numdisabled;
3669         int ret, i;
3670
3671         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3672         if (!rings)
3673                 return -ENOMEM;
3674
3675         for (i = 0; i < tun->numqueues; i++) {
3676                 tfile = rtnl_dereference(tun->tfiles[i]);
3677                 rings[i] = &tfile->tx_ring;
3678         }
3679         list_for_each_entry(tfile, &tun->disabled, next)
3680                 rings[i++] = &tfile->tx_ring;
3681
3682         ret = ptr_ring_resize_multiple(rings, n,
3683                                        dev->tx_queue_len, GFP_KERNEL,
3684                                        tun_ptr_free);
3685
3686         kfree(rings);
3687         return ret;
3688 }
3689
3690 static int tun_device_event(struct notifier_block *unused,
3691                             unsigned long event, void *ptr)
3692 {
3693         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3694         struct tun_struct *tun = netdev_priv(dev);
3695         int i;
3696
3697         if (dev->rtnl_link_ops != &tun_link_ops)
3698                 return NOTIFY_DONE;
3699
3700         switch (event) {
3701         case NETDEV_CHANGE_TX_QUEUE_LEN:
3702                 if (tun_queue_resize(tun))
3703                         return NOTIFY_BAD;
3704                 break;
3705         case NETDEV_UP:
3706                 for (i = 0; i < tun->numqueues; i++) {
3707                         struct tun_file *tfile;
3708
3709                         tfile = rtnl_dereference(tun->tfiles[i]);
3710                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3711                 }
3712                 break;
3713         default:
3714                 break;
3715         }
3716
3717         return NOTIFY_DONE;
3718 }
3719
3720 static struct notifier_block tun_notifier_block __read_mostly = {
3721         .notifier_call  = tun_device_event,
3722 };
3723
3724 static int __init tun_init(void)
3725 {
3726         int ret = 0;
3727
3728         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3729
3730         ret = rtnl_link_register(&tun_link_ops);
3731         if (ret) {
3732                 pr_err("Can't register link_ops\n");
3733                 goto err_linkops;
3734         }
3735
3736         ret = misc_register(&tun_miscdev);
3737         if (ret) {
3738                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3739                 goto err_misc;
3740         }
3741
3742         ret = register_netdevice_notifier(&tun_notifier_block);
3743         if (ret) {
3744                 pr_err("Can't register netdevice notifier\n");
3745                 goto err_notifier;
3746         }
3747
3748         return  0;
3749
3750 err_notifier:
3751         misc_deregister(&tun_miscdev);
3752 err_misc:
3753         rtnl_link_unregister(&tun_link_ops);
3754 err_linkops:
3755         return ret;
3756 }
3757
3758 static void __exit tun_cleanup(void)
3759 {
3760         misc_deregister(&tun_miscdev);
3761         rtnl_link_unregister(&tun_link_ops);
3762         unregister_netdevice_notifier(&tun_notifier_block);
3763 }
3764
3765 /* Get an underlying socket object from tun file.  Returns error unless file is
3766  * attached to a device.  The returned object works like a packet socket, it
3767  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3768  * holding a reference to the file for as long as the socket is in use. */
3769 struct socket *tun_get_socket(struct file *file)
3770 {
3771         struct tun_file *tfile;
3772         if (file->f_op != &tun_fops)
3773                 return ERR_PTR(-EINVAL);
3774         tfile = file->private_data;
3775         if (!tfile)
3776                 return ERR_PTR(-EBADFD);
3777         return &tfile->socket;
3778 }
3779 EXPORT_SYMBOL_GPL(tun_get_socket);
3780
3781 struct ptr_ring *tun_get_tx_ring(struct file *file)
3782 {
3783         struct tun_file *tfile;
3784
3785         if (file->f_op != &tun_fops)
3786                 return ERR_PTR(-EINVAL);
3787         tfile = file->private_data;
3788         if (!tfile)
3789                 return ERR_PTR(-EBADFD);
3790         return &tfile->tx_ring;
3791 }
3792 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3793
3794 module_init(tun_init);
3795 module_exit(tun_cleanup);
3796 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3797 MODULE_AUTHOR(DRV_COPYRIGHT);
3798 MODULE_LICENSE("GPL");
3799 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3800 MODULE_ALIAS("devname:net/tun");